an empirical investigation of wage discrimination in ... · jevili platovou diskriminaci v˚uˇci...
TRANSCRIPT
![Page 1: An Empirical Investigation of Wage Discrimination in ... · jevili platovou diskriminaci v˚uˇci Afroameriˇcan˚um a hr´aˇc˚um z Latinsk´e Ameriky v nejniˇzˇs´ım decilu](https://reader030.vdocument.in/reader030/viewer/2022040907/5e7d12e667f4fa2fa646aac6/html5/thumbnails/1.jpg)
Charles University
Faculty of Social SciencesInstitute of Economic Studies
BACHELOR THESIS
An Empirical Investigation of WageDiscrimination in Professional Football
Author: Jakub Blaha
Supervisor: Ing. David Kocourek
Academic Year: 2016/2017
![Page 2: An Empirical Investigation of Wage Discrimination in ... · jevili platovou diskriminaci v˚uˇci Afroameriˇcan˚um a hr´aˇc˚um z Latinsk´e Ameriky v nejniˇzˇs´ım decilu](https://reader030.vdocument.in/reader030/viewer/2022040907/5e7d12e667f4fa2fa646aac6/html5/thumbnails/2.jpg)
Declaration of Authorship
The author hereby declares that he compiled this thesis independently, using
only the listed resources and literature.
The author grants to Charles University permission to reproduce and to dis-
tribute copies of this thesis document in whole or in part.
Prague, July 26, 2017Signature
![Page 3: An Empirical Investigation of Wage Discrimination in ... · jevili platovou diskriminaci v˚uˇci Afroameriˇcan˚um a hr´aˇc˚um z Latinsk´e Ameriky v nejniˇzˇs´ım decilu](https://reader030.vdocument.in/reader030/viewer/2022040907/5e7d12e667f4fa2fa646aac6/html5/thumbnails/3.jpg)
Acknowledgments
Firstly, I would like to express my sincere gratitude to my supervisor Ing.
David Kocourek for his patience, motivation and expert knowledge. His guid-
ance helped during the whole duration of writing this thesis.
Besides my supervisor, I would like to thank to my parents, Renata and
Stanislav, and the rest of my family for the continuous mental and financial
support.
![Page 4: An Empirical Investigation of Wage Discrimination in ... · jevili platovou diskriminaci v˚uˇci Afroameriˇcan˚um a hr´aˇc˚um z Latinsk´e Ameriky v nejniˇzˇs´ım decilu](https://reader030.vdocument.in/reader030/viewer/2022040907/5e7d12e667f4fa2fa646aac6/html5/thumbnails/4.jpg)
Abstract
Salary discrimination is a phenomenon that arises from ineffective behaviour
of economic subjects. Even though its presence is incompatible with the the-
ory of profit maximization, salary inequality still persists in the human society.
Nevertheless, the investigation of this topic has been largely unheeded in the
environment of professional football. In our empirical research, we use the most
recent data to investigate the salary gap between white, African American and
Hispanic players in the American Major League Soccer. Besides ordinary least
squares method that focuses on the impact of ethnicity for the average player,
we adopted the method of quantile regression to reveal wage gap between play-
ers with below-average pays. Observing each player’s performance for 3 seasons,
we uncovered salary discrimination against African Americans and Hispanics
in the lowest decile of the salary distribution that amounts to 18.9% and 15.3%,
respectively. Furthermore, we utilized the difference-in-differences (DID) esti-
mator to find no effect of the increasing level of invested money on the wage
gap.
JEL Classification J30, Z20, Z21, J71 J31 J15
Keywords discrimination, race inequality, football, quan-
tile regression, OLS, wages, racism
Author’s e-mail [email protected]
Supervisor’s e-mail [email protected]
![Page 5: An Empirical Investigation of Wage Discrimination in ... · jevili platovou diskriminaci v˚uˇci Afroameriˇcan˚um a hr´aˇc˚um z Latinsk´e Ameriky v nejniˇzˇs´ım decilu](https://reader030.vdocument.in/reader030/viewer/2022040907/5e7d12e667f4fa2fa646aac6/html5/thumbnails/5.jpg)
Abstrakt
Platova diskriminace je jev, ktery je vysledkem neefektivnıho jednanı eko-
nomickych subjektu. Ackoli je jejı prıtomnost neslucitelna s teoriı maximalizace
zisku, platova nerovnost i presto nadale pretrvava v lidske spolecnosti. Nicmene,
pokud se jedna o prostredı profesionalnıho fotbalu, otazce platove diskrimi-
nace byva zrıdkakdy venovana pozornost. V nası studii pouzıvame nejnovejsı
data, abychom vysetrili rozdıly v platech mezi bılymi, afroamerickymi a latin-
skoamerickymi hraci v americke Major League Soccer. Krome metody nej-
mensıch ctvercu, ktera se zameruje na vliv rasy pro prumerneho hrace, jsme
pouzili kvantilovou regresi, abychom odhalili rozdıly v platu pro hrace s podpru-
mernou mzdou. Sledovanım statistik kazdeho hrace po dobu 3 sezon jsme ob-
jevili platovou diskriminaci vuci Afroamericanum a hracum z Latinske Ameriky
v nejnizsım decilu platoveho rozdelenı, ktera vysplhala k 18.9% a 15.3% ve
prospech bılych hracu. Mimoto jsme pouzili metodu difference-in-differences
(DID), abychom overili, ze stoupajıcı vyse investovanych penez nemela na rozdıl
v platech naprıc rasami zadny vliv.
Klasifikace JEL J30, Z20, Z21, J71 J31 J15
Klıcova slova diskriminace, rasova nerovnost, fotbal,
kvantilova regrese, OLS, platy, rasismus
E-mail autora [email protected]
E-mail vedoucıho prace [email protected]
![Page 6: An Empirical Investigation of Wage Discrimination in ... · jevili platovou diskriminaci v˚uˇci Afroameriˇcan˚um a hr´aˇc˚um z Latinsk´e Ameriky v nejniˇzˇs´ım decilu](https://reader030.vdocument.in/reader030/viewer/2022040907/5e7d12e667f4fa2fa646aac6/html5/thumbnails/6.jpg)
Contents
List of Tables viii
List of Figures ix
Acronyms x
Thesis Proposal xi
1 Introduction 1
2 Theory and Literature Review 5
2.1 Discrimination . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.2 Racial Discrimination . . . . . . . . . . . . . . . . . . . . . . . . 6
2.3 Wage Differential . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.3.1 Stastistical Discrimination . . . . . . . . . . . . . . . . . 8
2.3.2 Taste-based Discrimination . . . . . . . . . . . . . . . . . 8
2.4 Literature Review . . . . . . . . . . . . . . . . . . . . . . . . . . 10
3 Methodology and Crucial Terms 14
3.1 Independently pooled cross section . . . . . . . . . . . . . . . . 14
3.2 Difference-in-differences . . . . . . . . . . . . . . . . . . . . . . . 14
3.3 Quantile Regression . . . . . . . . . . . . . . . . . . . . . . . . . 16
4 League Overview and Dataset Construction 17
4.1 American Major League Soccer . . . . . . . . . . . . . . . . . . 17
4.2 Salary system . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
4.3 Dataset Construction . . . . . . . . . . . . . . . . . . . . . . . . 20
4.3.1 Performance statistics . . . . . . . . . . . . . . . . . . . 26
4.3.2 Other Factors . . . . . . . . . . . . . . . . . . . . . . . . 29
![Page 7: An Empirical Investigation of Wage Discrimination in ... · jevili platovou diskriminaci v˚uˇci Afroameriˇcan˚um a hr´aˇc˚um z Latinsk´e Ameriky v nejniˇzˇs´ım decilu](https://reader030.vdocument.in/reader030/viewer/2022040907/5e7d12e667f4fa2fa646aac6/html5/thumbnails/7.jpg)
Contents vii
5 Model 31
5.1 General Model . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
5.2 Our Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
6 Results 34
6.1 Discussion of the results . . . . . . . . . . . . . . . . . . . . . . 35
7 Conclusion 43
Bibliography 49
A OLS Assumptions I
B Robust methods IV
C Unrestricted vs. restricted model VI
![Page 8: An Empirical Investigation of Wage Discrimination in ... · jevili platovou diskriminaci v˚uˇci Afroameriˇcan˚um a hr´aˇc˚um z Latinsk´e Ameriky v nejniˇzˇs´ım decilu](https://reader030.vdocument.in/reader030/viewer/2022040907/5e7d12e667f4fa2fa646aac6/html5/thumbnails/8.jpg)
List of Tables
4.1 The highest paid MLS footballers in 2016 . . . . . . . . . . . . . 19
4.2 Occurrence in the dataset . . . . . . . . . . . . . . . . . . . . . 21
4.3 Definitions of variables . . . . . . . . . . . . . . . . . . . . . . . 24
4.4 Salary - Descriptive statistics . . . . . . . . . . . . . . . . . . . 24
4.5 Comparison between whole sample and lower quartile . . . . . . 25
4.6 Sample means, medians & standard deviations - 1 season . . . . 28
4.7 Sample means, medians & standard deviations - 3 seasons . . . 29
6.1 Relationship between race and position . . . . . . . . . . . . . . 35
6.2 OLS regressions of 3-seasons statistics . . . . . . . . . . . . . . . 39
6.3 OLS regressions of 1-season statistics . . . . . . . . . . . . . . . 40
6.4 Results of Quantile Regression . . . . . . . . . . . . . . . . . . . 41
A.1 Correlation of explanatory variables . . . . . . . . . . . . . . . . II
B.1 Comparison between OLS and White’s standard errors . . . . . IV
B.2 Comparison between OLS and Robust regressions . . . . . . . . V
C.1 Results after the exclusion of insignificant variables . . . . . . . VI
![Page 9: An Empirical Investigation of Wage Discrimination in ... · jevili platovou diskriminaci v˚uˇci Afroameriˇcan˚um a hr´aˇc˚um z Latinsk´e Ameriky v nejniˇzˇs´ım decilu](https://reader030.vdocument.in/reader030/viewer/2022040907/5e7d12e667f4fa2fa646aac6/html5/thumbnails/9.jpg)
List of Figures
4.1 MLS average salary of 100 best players . . . . . . . . . . . . . . 20
4.2 Development of players’ median salaries . . . . . . . . . . . . . . 21
4.3 3-seasons average statistics . . . . . . . . . . . . . . . . . . . . . 27
6.1 Correlation between GS and MINS . . . . . . . . . . . . . . . . 36
6.2 Quantile regression coefficients . . . . . . . . . . . . . . . . . . . 42
A.1 Q-Q plot of the regression . . . . . . . . . . . . . . . . . . . . . III
![Page 10: An Empirical Investigation of Wage Discrimination in ... · jevili platovou diskriminaci v˚uˇci Afroameriˇcan˚um a hr´aˇc˚um z Latinsk´e Ameriky v nejniˇzˇs´ım decilu](https://reader030.vdocument.in/reader030/viewer/2022040907/5e7d12e667f4fa2fa646aac6/html5/thumbnails/10.jpg)
Acronyms
FIFA Federation Internationale de Football Association
MLS Major League Soccer
OLS Ordinary Least Squares
IFAB International Football Association Board
GSSS Global Sports Salary Survey
MLB Major League Baseball
US United States
QR Quantile Regression
NHL National Hockey League
NBA National Basketball Association
FFP Financial Fair Play Regulations
DID Difference-in-differences
GAM General Allocation Money
![Page 11: An Empirical Investigation of Wage Discrimination in ... · jevili platovou diskriminaci v˚uˇci Afroameriˇcan˚um a hr´aˇc˚um z Latinsk´e Ameriky v nejniˇzˇs´ım decilu](https://reader030.vdocument.in/reader030/viewer/2022040907/5e7d12e667f4fa2fa646aac6/html5/thumbnails/11.jpg)
Bachelor Thesis Proposal
Author Jakub Blaha
Supervisor Ing. David Kocourek
Proposed topic An Empirical Investigation of Wage Discrimination in
Professional Football
Topic characteristics: To eradicate discrimination, intolerance and racism
from American soccer, Major League Soccer (MLS) joined Federation Inter-
nationale de Football Association (FIFA) in the campaign called ”Say no to
racism” (Mlssoccer.com 2010). On the occasion of Say No to Racism Day,
MLS stadiums incorporated field boards with “Say no to racism” signs. The
Commissioner of MLS Don Garber commented on this event as follows: ”With
players born in 44 different countries, our league reflects the broad range of eth-
nicities that love the game of soccer and that live in the United States.” Then
he added: ”Major League Soccer has a no-tolerance policy toward racism. The
league prohibits and will not tolerate any discrimination or harassment on the
basis of race.” After this statement, one might wonder if club owners and foot-
ball managers also comprehend the MLS’s fight against racial inequality.
In our thesis we would like to examine whether there is a presence of racial
discrimination among footballers from top American clubs and therefore to test
the effectiveness of MLS’s approach. As a measure of discrimination in Ameri-
can soccer, we investigate the relationship between the race and the footballer’s
income. We also take into account the fact that MLS eased the policy of salary
cap and established several policies such as the Designated Player Rule which
guarantees that the wages of the most productive players are solely determined
by the market forces. We incorporate this system shocks into our examining
so that the results obtained from our research would help us to answer the
![Page 12: An Empirical Investigation of Wage Discrimination in ... · jevili platovou diskriminaci v˚uˇci Afroameriˇcan˚um a hr´aˇc˚um z Latinsk´e Ameriky v nejniˇzˇs´ım decilu](https://reader030.vdocument.in/reader030/viewer/2022040907/5e7d12e667f4fa2fa646aac6/html5/thumbnails/12.jpg)
Bachelor Thesis Proposal xii
following research questions:
1. Do teams in Major League Soccer pay less to African American and
Hispanic players?
2. Is there a pay premium for white players in any part of the salary distri-
bution?
3. Did the change in wage policies worsen or improve the position of His-
panics and African Americans in terms of wages?
4. Are players racially discriminated across different positions?
Methodology This thesis uses econometric approach which is called Pooled
Cross Sections to estimate the model with factors influencing footballers’ in-
come. To estimate Pooled Cross Sections, we utilize the ordinary least squares
estimator, which is the most efficient one under given assumptions that need
to be satisfied. In addition, we incorporated the quantile regression to be able
to obtain estimates at different quantiles of players’ salary.
Outline
1. Introduction
2. Theory and Literature Review
3. Methodology and Crucial Terms
4. Dataset Construction and League Overview
5. Model
6. Results
7. Conclusion
Core bibliography
1. Arrow, Kenneth J. (1998): “What Has Economics to Say About Racial Discrimina-
tion?.” Journal of Economic Perspectives) pp. 91-100.
2. Fryer, Roland G (2010): “Racial inequality in the 21st century: The declining signifi-
ance of discrimination.” National bureau of economic research - Harvard University
3. KAHN, Lawrence M & K. LANG (2000): “The Sports Business as a Labor Market
Laboratory.” Journal of Economic Perspectives. American Economic Association
4. LEHMANN, Jee-Yeon K. & K. LANG (2011): “Racial Discrimination in the Labor
Market: Theory and Empirics.” Journal of Economic Literature pp. 959- 1006.
5. WOOLDRIDGE, Jeffrey M., (2006): “Introductory econometrics: a modern ap-
proach. .” 3rd ed. Mason: Thomson/South-Western
![Page 13: An Empirical Investigation of Wage Discrimination in ... · jevili platovou diskriminaci v˚uˇci Afroameriˇcan˚um a hr´aˇc˚um z Latinsk´e Ameriky v nejniˇzˇs´ım decilu](https://reader030.vdocument.in/reader030/viewer/2022040907/5e7d12e667f4fa2fa646aac6/html5/thumbnails/13.jpg)
Bachelor Thesis Proposal xiii
Author Supervisor
![Page 14: An Empirical Investigation of Wage Discrimination in ... · jevili platovou diskriminaci v˚uˇci Afroameriˇcan˚um a hr´aˇc˚um z Latinsk´e Ameriky v nejniˇzˇs´ım decilu](https://reader030.vdocument.in/reader030/viewer/2022040907/5e7d12e667f4fa2fa646aac6/html5/thumbnails/14.jpg)
Chapter 1
Introduction
”The only difference between man and man all the world over is one of degree,
and not of kind...Where is the cause for anger, envy or discrimination?”
Mahatma Gandhi
Association football, commonly known as football1, is a sport where each of
both teams consist of 11 players and the main objective of each side is simple -
score as many goals as possible to the opponent in order to win the game. Not
surprisingly, it is the most popular sport in the entire world (Dunning et al.
1999) as there are approximately 265 million registered players (Fifa.com 2007)
in 200 dependencies. The rules of the game were determined by International
Football Association Board (IFAB), which was established in 1886 and ever
since, the money involved in this sport has skyrocketed.
Nonetheless, the labour market of football players is nowadays a highly-
competitive one since out of the millions of football players only a fraction can
make a living by playing football. The increasing interest of the media and the
leading sports companies caused that the money invested in it has increased ex-
ponentially as football was more and more professionalised. This phenomenon
can be seen for example in terms of wages of footballers which have augmented
dramatically. For comparison, in 1901 the Football League in England imposed
a restriction that no player would be paid more than £4-a-week2 for his per-
formance. According to Sportingintelligence.com (2016), though, the average
1Even though the name ”football” might signify American football, we decided to use thisword throughout this paper as it is more widespread in Europe.
2If we adjust for inflation, this value would yield approximately £443.14 in today´s money(May, 2017).
![Page 15: An Empirical Investigation of Wage Discrimination in ... · jevili platovou diskriminaci v˚uˇci Afroameriˇcan˚um a hr´aˇc˚um z Latinsk´e Ameriky v nejniˇzˇs´ım decilu](https://reader030.vdocument.in/reader030/viewer/2022040907/5e7d12e667f4fa2fa646aac6/html5/thumbnails/15.jpg)
1. Introduction 2
Premier League player earned approximately £49,211 a week.
From an employer’s point of view, the market of football players is a spe-
cific type of labour market. On the one hand, it differs from other working
environments in the possibility to perfectly observe worker’s performance us-
ing statistical data. On the other hand, it presumably shares some similar
characteristics with other types of markets with employees, such as observable
elements of supply and demand (Rosen & Sanderson 2001), which need to be
taken into consideration. Owing to the publication of the Civil Rights Act of
1964, the most discussed issue in neoclassical era is the fact that different groups
of employees, be they skilled or not, receive different wages (Wells et al. 2001).
This treatment of making a distinction against a person based on their age,
race or gender is called discrimination and its presence provoked response in all
kinds of social affairs (Arrow 1998). Since it represents a specific sort of eco-
nomic inefficiencies, it is often a subject of economic academic literature. Being
perceived as socially unacceptable and hostile, racial discrimination has been
probably one of the most considered forms due to the fact that various policies
have been made to remove the difference in dissimilar treatment through races.
The motivation for these policies has been here since the colonial era, because
European Americans have been given exclusive privileges in terms of voting
rights, education, citizenship and most importantly for our study, better work-
ing conditions including higher wages (Sears 1988). Nevertheless, the definition
of discrimination states that there is no presence of disriminatory behaviour if
the persons observed do not have the same productivity. However, productivity
arises from such factors as education or training that have been provided in a
different way to people based on their race. However, the professional football
is a very unique case of labour market. First of all, this sports industry is
under general public scrutiny and players should be treated the same in terms
of career development. Second, clubs are disclosed to the performance of their
employees as they are provided with detailed game statistics. Hence, football
clubs have complete knowledge about the productivity of their employees. In
connection with this fact, wage discrimination should not be enabled and wage
differences should only reflect the importance of a player for the team. Thus,
if salary discrimination could be detected in such a competitive, visible and
scrutinized industry, the results of the research might encourage academics to
study discrimination also in other areas.
![Page 16: An Empirical Investigation of Wage Discrimination in ... · jevili platovou diskriminaci v˚uˇci Afroameriˇcan˚um a hr´aˇc˚um z Latinsk´e Ameriky v nejniˇzˇs´ım decilu](https://reader030.vdocument.in/reader030/viewer/2022040907/5e7d12e667f4fa2fa646aac6/html5/thumbnails/16.jpg)
1. Introduction 3
One of the reasons for clubs to have nondiscriminatory behaviour is the
fact that they can be considered as profit maximisers (Zimbalist 2003), and
therefore they should set the marginal revenue equal to the marginal cost in
order to attain the highest profit. Speaking in general terms, the worker should
receive the very same amount of money as he is capable of earning for his em-
ployer. The question that might arise is whether the clubs remunerate their
players justly based on their perfomance or they use prejudices against, or in
favour of players with non-white origins. There have been many studies ex-
amining the relationship between race and compensation in sport, namely in
professional basketball and baseball, but only a few were devoted to football
due to the unavailability of data. Szymanski (2000) uses wage expenditure and
performance statistics in order to test the presence of racial inequality among
top English clubs. Discovering positive ceteris paribus effect of the proportion
of black players employed on the team performance, he proposed a discrimina-
tory behaviour from some team owners who preferred purchasing white players
in spite of worse results. A direct effect of race on player’s compensation in
National Basketball Association was investigated by Hamilton (1997). Once
controlled for performance, he devoted his research to salary differential and
compared the situation in mid-1980s and 1990s. Controlling for players’ char-
acteristics, he used OLS and Tobit regressions to come to the conclusion that
20% premium paid to white players in 1985 were removed after one decade.
However, once examining the same dataset by censored quantile regression, he
discovered two curiosities. In the lower end of the dataset, the whites earned
substantially less than their black counterparts. But, the premium of 18% was
received by whites in the upper end of the distribution. The outcome of his
study is very unexpected as discriminating better and expensive players tends
to be more costly for the employer. His results were reconfirmed by the paper
of Kahn & Shah (2005) who also found wage discrimination against non-white
players. These findings lead to a question, whether wages of white and non-
white players also differ in football - the most followed and commercialized
sport in the world (Mueller et al. 1996). Even though Major League Soccer
would appear as an improbable place to detect racial discrimination, no study
has been focused on this topic in recent years.
Our thesis aims to clarify two ambiguities. First of all, we investigate the
relationship between players’ income and race in Major League Soccer, which is
the only national league that completely discloses the information about play-
![Page 17: An Empirical Investigation of Wage Discrimination in ... · jevili platovou diskriminaci v˚uˇci Afroameriˇcan˚um a hr´aˇc˚um z Latinsk´e Ameriky v nejniˇzˇs´ım decilu](https://reader030.vdocument.in/reader030/viewer/2022040907/5e7d12e667f4fa2fa646aac6/html5/thumbnails/17.jpg)
1. Introduction 4
ers’ salaries3 and at the same time we utilize the American data that was used
in the majority of researches on this topic. We are highly inspired by the most
recent studies and besides classical OLS method, we also adopt the quantile
regression analysis to be able to investigate racial discrimination in different
quantiles of the salary. Aditionally, we aim to find out the effect of several
salary policies that were introduced to augment the quality of the league. As
an example we mention the Designated Player Rule that allowed MLS clubs to
sign players who would otherwise be considered above their salary cap. This
approach will provide us with a knowledge, whether or not such a system shock
caused a wage gap between Hispanic, African American and white players. The
structure of the thesis looks as follows:
Chapter 2 summarizes the literature written about the racial inequality in
sport and other labour markets. In addition, we provide the theory standing
behind the inequity and intolerance in working process - crucial conception for
our thesis. The forms of discrimination are mentioned as well. Chapter 3 pro-
vides an introduction to pooled independently cross-sections data estimation
and explains the purpose of the Difference-in-Differences estimator. Last but
not least, we present the quantile regression to be able to examine individual
quantiles of the salary distribution. Chapter 4 deals with the dataset con-
struction and also introduces plenty of descriptive statistics so that the reader
has better notion about the distribution of the data. Moreover, it covers the
description of the Major League Soccer which is the field of our exploration.
Chapter 5 describes the model used for answering the research questions. The
discussion of results from our regressions is presented in Chapter 6. We also
find it necessary to point out some shortcomings of the investigation. Finally,
Chapter 7 summarizes our findings and acts as a spur for further research.
3As of June 2017, the data was available at: mlsplayers.org
![Page 18: An Empirical Investigation of Wage Discrimination in ... · jevili platovou diskriminaci v˚uˇci Afroameriˇcan˚um a hr´aˇc˚um z Latinsk´e Ameriky v nejniˇzˇs´ım decilu](https://reader030.vdocument.in/reader030/viewer/2022040907/5e7d12e667f4fa2fa646aac6/html5/thumbnails/18.jpg)
Chapter 2
Theory and Literature Review
This chapter explains the theory standing behind the economics of discrimi-
nation. In the first part, we define what discrimination is, we also discuss its
forms and consequences. In the second part we introduce the principle of the
wage gap and two types of wage discrimination. In the third part, the reader
is informed about the literature written about this phenomenon and we also
present the results of previous studies in sports industry.
2.1 Discrimination
The word discrimination comes from the latin verb discrimire which signifies
”to separate, to make a distinction” (Oxford University 2017). It can therefore
be assumed that discrimination is a treatment of making a distinction against,
or in favor based on the group where the given person pertains. This act sig-
nificantly contradicts essential principles of human rights and represents one
of the strongest economic inefficiencies. There are different types of discrimi-
nation, though. Unequal treatment can be aimed at persons of different age,
race, religion or nationality. Since the main objective of the study is dissimilar
treatment based on the person’s ethinicity, this thesis primarily concerns racial
discrimination and its direct impacts.
The key assumption when considering a different treatment by employers
based on the employees’ race is that we should observe people of equal pro-
ductivity (Becker 1971). This condition is hardly measured through labour
market as employers seldom have perfect information about the performance
of their employees and this limitation markedly reduces the possibilities for po-
![Page 19: An Empirical Investigation of Wage Discrimination in ... · jevili platovou diskriminaci v˚uˇci Afroameriˇcan˚um a hr´aˇc˚um z Latinsk´e Ameriky v nejniˇzˇs´ım decilu](https://reader030.vdocument.in/reader030/viewer/2022040907/5e7d12e667f4fa2fa646aac6/html5/thumbnails/19.jpg)
2. Theory and Literature Review 6
tential investigations. There are still some researches dealing with employment
discrimination which can be measured, for example, at the moment of initial
hire. One of them was introduced by Pager et al. (2009) who found patterns
between the probability of being hired and demographic characteristics in the
US. According to the research he made, Hispanic and black applicants with
clean backgrounds had approximately the same chances of being hired as white
applicants recently released from prison. Interpersonal skills of applicants are
arguable but for the sake of justifiability, the candidates for job offers were
given identical resumes.
However, the crucial statement that should be noted is that the environment
of professional sport has one special feature which distinguishes it from the rest
of the labour market. The performance and productivity of its participants
are measurable since the statistics of each individual are observed and can be
subsequently used for research studies. Thanks to this, we are able to obtain
all data needed to determine whether the difference in terms of wages also
exists between African American, Hispanic and white footballers. But first, we
find it necessary to provide some key facts about the situation in the American
market with workers as racism has been rooted in the American society since the
establishment of the country (Takaki 1979). Hence, we find it very important
for reader’s perception about the meaning of our research.
2.2 Racial Discrimination
Racial discrimination remains present in every aspect of human society (Arrow
1998). Fortunately, in the course of 20th century, life of the blacks improved
dramatically. The difference in unequal treatment in United States was in-
tended to be abolished in 1964 when Civil Rights Act was enforced. It ended
racial separation in schools, prohibition on voting and most importantly, worse
conditions at the workplace. Besides, the US government implemented a policy
which is known as affirmative action to assure that groups like African Amer-
icans, who had been a subject to discrimination for ages, would be accepted
more often when applying for a job (Feinberg 2003). Not surprisingly, these
political acts motivated plenty of researchers to scrutinize racism and its con-
sequences in all areas of labour market (Cain 1986).
Even though racial inequality is not as prevalent nowadays as it was earlier
![Page 20: An Empirical Investigation of Wage Discrimination in ... · jevili platovou diskriminaci v˚uˇci Afroameriˇcan˚um a hr´aˇc˚um z Latinsk´e Ameriky v nejniˇzˇs´ım decilu](https://reader030.vdocument.in/reader030/viewer/2022040907/5e7d12e667f4fa2fa646aac6/html5/thumbnails/20.jpg)
2. Theory and Literature Review 7
in the past, its continous existence still can be proved in many aspects of social
affairs. Most importantly, racial discrimination can be also found in economic
aspects, namely, in terms of income, life expectancy or hiring standards. In the
United States, blacks earn 24% less compared to whites and live on average
5 fewer years. Concerning Hispanics, they earn 25% less than whites hold-
ing other factors fixed (Fryer Jr 2010). This empirical fact inspired many to
examine conditions in professional team sports, especially in North America.
Although racial inequality produces a certain opportunity cost, it still exists in
many sports industries (see section 2.4). However, before we introduce the find-
ings of racial discrimination in sport, the section 2.3 explains two behaviours
that can lead to the race pay gap.
2.3 Wage Differential
Wage discrimination in the labour market occurs if, after controlling for special
characteristics such as experience or education, the coefficient on race, gender
or age is negative and statistically significant. One of the first studies that took
wage gap into consideration was introduced by Oaxaca (1973). He inquired into
the wage differential between genders and he clarified that male-female wage
difference is not caused by discrimination, but it rather stems from the vari-
ance in productivity. The most important limitation of his study lies in the
fact that he did not have an adequate amount of data at his disposal. He
was consequently followed by researchers who introduced various extensions
in order to provide results of higher importance. Subsequent studies included
also time dimension into their models allowing them to measure returns to
skills, and how they have been changing the wage gap over time (Blau & Kahn
2016). This technique identifies whether wage difference among people of dif-
ferent ethnicity has widened or narrowed, and if the factors that are hidden
in the racial dummy variable, have augmented or lowered their effects over time.
As the studies dealing with racial discrimination have been accumulating,
literature has been divided into two branches. The first is statistical and it re-
lies on the assumption that the employer would lean his decision on the group
averages if they did not have perfect information about employees. Taste-based
discrimination, by contrast, is based on the prejudice of decision-maker. The
second appears to be more likely our case since we are able to observe the per-
![Page 21: An Empirical Investigation of Wage Discrimination in ... · jevili platovou diskriminaci v˚uˇci Afroameriˇcan˚um a hr´aˇc˚um z Latinsk´e Ameriky v nejniˇzˇs´ım decilu](https://reader030.vdocument.in/reader030/viewer/2022040907/5e7d12e667f4fa2fa646aac6/html5/thumbnails/21.jpg)
2. Theory and Literature Review 8
formance of each player. Nevertheless, we explain the concepts of both of them.
2.3.1 Stastistical Discrimination
The pioneers of this form of discrimination were Edmund Phelps and Kenneth
Arrow (Fang & Moro 2010). The key assumption of this type of discrimination
is that the employer evaluates their workers without having complete knowl-
edge about their productivity. The only characterictics that can be observed by
the employer are gender, age, experience, ethnicity or race. Therefore, work-
ers can disclose their skills only by the form of resume and partly during the
process of interview. Otherwise, unobservable characterictics are presumed to
be correlated with observable features and workers are classified accordingly
(Arrow 1971). Statistical discrimination occurs when employers treat poten-
tial employees based on the group where they belong (Lang et al. 2012). The
question which might arise is whether the workforce productivity varies across
races. Altonji & Pierret (2001) came to the conclusion that race is a significant
variable influencing productivity of workers. The evidence of statistical dis-
crimination was found and moreover, they clarified that the more information
the employer obtains about their workers, the less significant easily observable
variables become as determinants of workers’ earnings. A model of statistical
discrimination has been used by Lang & Manove (2011) who examined a wage
differential between whites and blacks. As workers’ ability was unobserved,
they used the score on Armed Forces Qualification Test as a proxy for their
ability. Their study vindicated that the wage difference between white and
black workers is equal to the return of additional year of education, holding
other factors fixed.
2.3.2 Taste-based Discrimination
Principles of this theory were presented by Becker (1971). The crucial as-
sumption for his model is that there is perfect competitiveness among firms.
Furthermore, we need to assume that employers are perfectly informed about
workers’ race. While the latter condition is undoubtedly satisfied in profes-
sional football environment, the first mentioned might be debatable in the case
of Major League Soccer, since clubs following discriminatory behaviour, and
therefore accruing negative profits created by discriminating, cannot so easily
![Page 22: An Empirical Investigation of Wage Discrimination in ... · jevili platovou diskriminaci v˚uˇci Afroameriˇcan˚um a hr´aˇc˚um z Latinsk´e Ameriky v nejniˇzˇs´ım decilu](https://reader030.vdocument.in/reader030/viewer/2022040907/5e7d12e667f4fa2fa646aac6/html5/thumbnails/22.jpg)
2. Theory and Literature Review 9
leave the market in the long-run.
According to this theory, the utility of the employer is dependent on profit
(π) and on the amount of non-white workers (Lb). The former mentioned
influences the utility function positively whereas the latter negatively. As each
employer has different taste for prejudices, coefficient dp states the relationship
between utility and number of prejudiced workers.
ue = π − dp · Lb (2.1)
Equation 2.1 presumes that both whites and blacks have equal marginal
productivity implying that employers avoid cooperation with a specific group
of employees with the same characteristics no matter how productive they are.
The profit function therefore looks as follows:
π = f(Lw + Lb)− ww · Lw − wb · Lb (2.2)
where ww and wb stand for wages for black and white workers, Lw is an amount
of white workers and f is a production function. Microeconomics suggests that
employers would choose the number of white workers Lw, and the number of
black workers Lb that maximize the utility function:
ue = f(Lw + Lb)− ww · Lw − wb · Lb − dp · Lb
These values L∗b and L∗
w must satisfy the conditions given below, so-called first-
order conditions.
f ′(L∗w + L∗
b)− ww = 0, L∗w > 0 (2.3)
f ′(L∗w + L∗
b)− wb − dp = 0, L∗b > 0 (2.4)
These two conditions imply that employers are willing to hire new labour
force until the moment, when the marginal cost and the marginal product are
in equality. In case of white workers, the marginal cost is only ww. However, in
case of black workers, the marginal cost for employer is their wage wb, added to
the discrimination coefficient dp. Given the assumptions we have stated earlier,
solely white workers would be hired if ww − wb < dp. On the other hand, only
blacks would be employed if ww − wb > dp. We can easily deduce that the
market equilibrium occurs if
w∗w = w∗
b + d∗p (2.5)
![Page 23: An Empirical Investigation of Wage Discrimination in ... · jevili platovou diskriminaci v˚uˇci Afroameriˇcan˚um a hr´aˇc˚um z Latinsk´e Ameriky v nejniˇzˇs´ım decilu](https://reader030.vdocument.in/reader030/viewer/2022040907/5e7d12e667f4fa2fa646aac6/html5/thumbnails/23.jpg)
2. Theory and Literature Review 10
where w∗w and w∗
b are optimum wages. In the equilibrium, discrimination coef-
ficient d∗p of the marginal discriminator would be equal to the wage differential
between black and white workers. We can therefore conclude that the wage
differential is positively correlated with the prejudice of the marginal employer
and its magnitude also depends on his taste for discrimination.
2.4 Literature Review
The problem of racial discrimination has motivated plenty of economists to
investigate this social discrepancy. Gary Becker (1971) wrote his paper The
Economics of Discrimination where he divided discrimination into three main
types - employer, co-worker and customer. Each one has its name after the
group which tends to pursue discriminatory behaviour. He asserted that firms
that consider their workers as equal, and treat them so, have competitive ad-
vantage against the firms that discriminate, as they do not have to cope with
additional costs caused by unequal treatment. Discriminating firms are there-
fore forced to leave the market because of inefficiency. On the other hand,
avoiding discrimination among co-workers should result in equally segregated
teams that are chiefly highly-competitive. Lastly, customer discriminatory be-
haviour in the sports industry can be observed if supporters had trouble at-
tending matches where the opponent team woud be composed of more players
of a different race than usual.
As far as professional football is concerned, very few researches have been
written about and thus, we provide also results from another sports sectors.
One of the first studies that engaged in the topic of racism in sport was exe-
cuted by Gwartney & Haworth (1974), who, in their work focused on the Major
League Baseball (MLB), discovered patterns between the success rate and par-
ticipation of black players in a team. Their sample was compiled from games
played between 1947 and 1956. Statistics proved that five teams with highest
number of African Americans were among six teams with highest win per game
ratio. Another finding was introduced by Eitzen et al. (1982) who alleged that
professional sport can be viewed as a field of equal economic opportunity for
minorities since teams desire to maximize profit. Their conjecture that sports
provide an exceptional opportunity for minorities was bolstered by the fact that
representation of non-white players in major teams is much higher than in the
rest of labor force. According to Kahn (1991), 74.3% of all players were black
![Page 24: An Empirical Investigation of Wage Discrimination in ... · jevili platovou diskriminaci v˚uˇci Afroameriˇcan˚um a hr´aˇc˚um z Latinsk´e Ameriky v nejniˇzˇs´ım decilu](https://reader030.vdocument.in/reader030/viewer/2022040907/5e7d12e667f4fa2fa646aac6/html5/thumbnails/24.jpg)
2. Theory and Literature Review 11
in NBA during the season 1985-86. In MLB, the proportion was only 27.8%.
Still relevant, though, if we take into account the fact that the very first black
player who was selected to play alongside his white team-mates made his major
league debut on April 15, 1947. His name was Jackie Robinson and his story
even inspired the movie director Brian Helgeland to film a documentary about
him1. Thanks to his perseverance and bravery, he won recognition for his per-
formance even though he faced many types of discrimination. First of all, some
of his team-mates refused to play along his side. One member of his team even
required to be traded somewhere else rather than play with him in the same
team. In addition to that, he was constantly being humiliated by his opponents.
But, the management of Brooklyn Dodgers were convinced about him being in-
volved in the team and decided to penalize others rather than play against him.
In the paragraphs above, we covered examples of co-worker and employer
discrimination, customer discriminatory behaviour is much more usual than
these two, though. Unlike co-worker and employer prejudice, inequal treat-
ment from customers cannot be eliminated by market forces. According to
Scully (1974), presence of African Americans in MLB significantly descreased
team revenue, all else being equal. This finding probably reflects preferences of
white fans who might have preferred attending matches with non-black players.
Nevertheless, there was no evidence of lower revenues based on the racial com-
position of playing teams in the season 1976-77 (Sommers & Quinton 1982).
Brown et al. (1991) discovered a relation between fans and wage discrimina-
tion. According to their study, wage differential in Major League Baseball was
for the most part caused by the decision-making of spectators2. As a proof
of discriminatory behaviour we can also consider the finding of Nardinelli &
Simon (1990)3 who explored the market with players’ baseball cards. Cards
with white players were sold for higher prices than for equally qualified blacks.
This might suggest racial prejudice among fans, but the demand for baseball
cards can be completely different from that for matches.
Another form of discrimination was found by Jiobu (1988) who came to
the conclusion that, if controlled for performance, black players had a signifi-
1The name of the film is ”42”, which points to the number on his shirt.2In their study, they revealed an insignificant, negative effect on attendance. Replacing
one white player for black player resulted in 8,400 fans lost.3If we control for performance, black and Hispanic players had significantly lower prices
of baseball cards, the magnitude was -14.2% and -9.8%, respectively.
![Page 25: An Empirical Investigation of Wage Discrimination in ... · jevili platovou diskriminaci v˚uˇci Afroameriˇcan˚um a hr´aˇc˚um z Latinsk´e Ameriky v nejniˇzˇs´ım decilu](https://reader030.vdocument.in/reader030/viewer/2022040907/5e7d12e667f4fa2fa646aac6/html5/thumbnails/25.jpg)
2. Theory and Literature Review 12
cantly higher exit rate from Major League Baseball than whites. On the other
hand, the exit rate of Hispanics was not significantly different from zero during
the seasons 1971-1985. There exists also researches that investigate the game
statistics. Scully (1973) unreveals that African Americans have significantly
higher points per minute ratio in NBA. In his research through sport envi-
ronment he also spotted that blacks outperform whites in American football.
Their dominance was caught in all performance aspects and the difference was
found to be significant at 10 out of 12 variables. Kahn & Sherer (1988) revis-
ited the case of wage segregation. They compared players with same salaries
and concluded that at the same level of wage, blacks outperform white players4.
As was said earlier, race segregation can have many forms. One of the
biggest motivation for our thesis was the discovery of Holmes (2011). He used
quite nontraditional method that is called quantile regression to show signifi-
cant differences in individual points of the salary distribution of MLB players.
Even though he was inspired by Hamilton (1997), his study came to the exactly
opposite outcome. While Hamilton examined NBA and discovered 18% pre-
mium for whites5 in the upper part of the distribution, Holmes’ investigation
of MLB uncovered 25% pay gap against blacks in the lower quartile. His paper
serves as a proof that even though racial discrimination causes economic losses
and conflicts in society, it is still accepted by the general public, even in such
a scrutinized industry of professional sport.
In comparison with North America, research studies on the topic of racial
discrimination in European professional sports are extensively limited primarily
due to the lack of data. Wilson & Ying (2003) suggest that team performance in
Europe could be considerably improved by hiring players from Latin America.
Their work was based on the data from the period after the Bosman6 ruling
was enforced. According to Frick (2006) who surveyed the German Bundesliga,
4Nevertheless, there was no race difference concerning players being drafted by NBAteams.
5In his model, he did not include a covariate implying the player’s height. The results ofhis study might be biased as whites are on average higher than their black counterparts.
6Jean-Marc Bosman was a Belgian playing for RFC Liege. After his contract expired in1990, he wanted to change teams but his potential new team refused to accept his transferfee. His wage was afterwards reduced and he no longer played for the first team. Whichresulted in taking his case to the European Court of Justice. He ended up suing for restraintof trade. Afterwards, Bosman and all players in EU were enabled to transfer for free at theend of their existing contracts. Formerly, footballer were the property of their clubs. Butnow, they become employers like any other workers in European Union.
![Page 26: An Empirical Investigation of Wage Discrimination in ... · jevili platovou diskriminaci v˚uˇci Afroameriˇcan˚um a hr´aˇc˚um z Latinsk´e Ameriky v nejniˇzˇs´ım decilu](https://reader030.vdocument.in/reader030/viewer/2022040907/5e7d12e667f4fa2fa646aac6/html5/thumbnails/26.jpg)
2. Theory and Literature Review 13
players fromWestern Europe, Eastern Europe and Latin America receive higher
bonuses relative to their German peers. The differences were found to be
significant with values of 15% in favour of Western Europeans, 30% for players
from Eastern Europe and a premium of 50% for ”football artists” from Latin
America.
![Page 27: An Empirical Investigation of Wage Discrimination in ... · jevili platovou diskriminaci v˚uˇci Afroameriˇcan˚um a hr´aˇc˚um z Latinsk´e Ameriky v nejniˇzˇs´ım decilu](https://reader030.vdocument.in/reader030/viewer/2022040907/5e7d12e667f4fa2fa646aac6/html5/thumbnails/27.jpg)
Chapter 3
Methodology and Crucial Terms
The main task of this chapter is to present econometric concepts that we have
been using during the analysis for every step of the practical part to become
easier to comprehend.
3.1 Independently pooled cross section
Independently pooled cross section sample is obtained if we randomly draw
observations from each time period (Wooldridge 2015). This approach of data
collection has many reasons. Primarily, we have a larger sample which results
in more precise estimators. Furthermore, test statistics become more valid.
The other reason standing behind this approach is the fact, that the dependent
variable might differ just because of time. As the distributions of population
could be different across time periods, we include t−1 dummy variables, where t
is the number of time periods we observe. This measure avoids dummy variable
trap and also allows the intercept to vary across periods. If we interact a period
dummy variable with one of the explanatory variables included into the model,
we can assess if the impact of this covariate has changed over time.
3.2 Difference-in-differences
If we are interested in studying an impact of a policy, a law, or the effect
of a treatment, the method which is used in quantitative researches is called
difference-in-differences (Ashenfelter & Card 1984). To be able to apply difference-
in-differences (DID) estimator, we require data both from a control and treat-
![Page 28: An Empirical Investigation of Wage Discrimination in ... · jevili platovou diskriminaci v˚uˇci Afroameriˇcan˚um a hr´aˇc˚um z Latinsk´e Ameriky v nejniˇzˇs´ım decilu](https://reader030.vdocument.in/reader030/viewer/2022040907/5e7d12e667f4fa2fa646aac6/html5/thumbnails/28.jpg)
3. Methodology and Crucial Terms 15
ment group1 at least at two time periods. In a simplified form, we can write
the model as follows:
y = β0 + δ0T2 + β1GoI + δ1T2 ·GoI + u (3.1)
Where y is the response variable, T2 is a dummy implying that the data was
collected from the second time period and GoI is a dummy variable equal to 1,
if the observation belongs to the group of interest. The coefficient of interest
is in our case δ1, since it captures the effect of interaction term T2 · GoI. The
logic behind the difference-in-differences estimator is described below.
E[y | GoI, T2] = β0 + δ0 + β1 + δ1
E[y | OO, T2] = β0 + δ0
Where the abbreviation ”OO” stands for other observations that are not in-
cluded in the group of interest. T1 signifies the base year. If we subtract these
two equation from each other, we obtain the first difference:
E[∆yT2] = E[yGoI − yOO | T2] = β1 + δ1 (3.2)
The other difference is obtained if we subtract expected values from the base
year.
E[y | GoI, T1] = β0 + β1
E[y | OO, T1] = β0
And the difference is therefore:
E[∆yT1] = E[yGoI − yOO | T1] = β1 (3.3)
The final step of obtaining DID estimator looks as follows:
E[∆yT2−∆yT1
] = δ1 (3.4)
The coefficient δ1 will in our case reflect, how much the salaries of African
Americans and Hispanics have changed with the increasing commercialization
of MLS together with the introduction of new salary policies. As the first signif-
1So-called control group is the baseline measure for the experiment. On the other hand,the treatment group is the one that receives experimental manipulation.
![Page 29: An Empirical Investigation of Wage Discrimination in ... · jevili platovou diskriminaci v˚uˇci Afroameriˇcan˚um a hr´aˇc˚um z Latinsk´e Ameriky v nejniˇzˇs´ım decilu](https://reader030.vdocument.in/reader030/viewer/2022040907/5e7d12e667f4fa2fa646aac6/html5/thumbnails/29.jpg)
3. Methodology and Crucial Terms 16
icant increase in wages occured in 20102 (see Model 1 in the table 6.2), we inter-
act this time variable with race dummies to obtain estimates for y2010*Black
and y2010*Hisp. These dummies are intended to uncover whether an increase
in investment has had any discriminatory elements or not. If we observed a
change in the salaries of black and Latin players only, before and after the
easing of the salary cap, we would fail to control for some omitted variables,
such as economic situation in the US or popularity of soccer there. Once we
also use white players as a control in the DID model, we eliminate a possible
bias of our estimator, even though some of the variables influencing players’
salary are unobserved.
3.3 Quantile Regression
Quantile regression was introduced by Koenker & Bassett Jr (1978) and offers
another approach for data analysis. While OLS gives us estimates at the aver-
age, quantile regression accounts for outliers and can estimate any quantile q for
q ∈ (0, 1) of the explained variable. This treatment ensures higher robustness
of our estimates. Thanks to this tool, we are able to measure the effect of race
on compensation for less paid players that are more susceptible due to their
relative lower expenses for employers. One of the advantages might be also
that analysis of characteristics between explanatory and explained variables
becomes richer and more understandable as we are not focused only on the
conditional mean. The estimator of QR for quantile q minimizes the function:
Q(βq) =N∑
i:yi≥x′
iβ
q | yi − x′
iβq | +N∑
i:yi<x′
iβ
(1− q) | yi − x′
iβq | (3.5)
where q varies depending on the quantile we want to observe. The coefficient
interpretation remains the same as in the case of OLS, except, instead of the
average, we refer to the corresponding quantile. Another advantage seems to be
the fact that quantile regression does not require homoskedasticity and normal
distribution of the error terms. These two assumptions are rarely satistified
in the case of the salary distribution of the entire population because of high
extreme values. For more detailed description of this method, see Koenker &
Hallock (2001).
2In our own interest, we tried to interact all of the season dummies, but none of them hasbeen found significant for the model.
![Page 30: An Empirical Investigation of Wage Discrimination in ... · jevili platovou diskriminaci v˚uˇci Afroameriˇcan˚um a hr´aˇc˚um z Latinsk´e Ameriky v nejniˇzˇs´ım decilu](https://reader030.vdocument.in/reader030/viewer/2022040907/5e7d12e667f4fa2fa646aac6/html5/thumbnails/30.jpg)
Chapter 4
League Overview and Dataset
Construction
This part of our thesis is dedicated to a detailed description of the data used in
our survey. First of all, we decided to provide some basic information about the
MLS as well as to highlight its financial standing. The Major League Soccer
was chosen for our analysis due to its uniqueness in terms of publishing players’
salaries. Another reason was the fact that it reflects the typical elements of
North American sports to which most of the studies concerning discrimination
in professional sports have been devoted. The next step of this chapter is a
description of some policies that helped MLS to increase its quality and to
adjust the salary system that was introduced with establishment of the league.
The most famous one is the Designated Player Rule that was established in
2007 in order to allow clubs to pay transfers and wages exceeding the salary
cap. Last part of this chapter focuses on the detailed description of the dataset
and to the justification of each variable used for estimating the determinants
of player’s compensation.
4.1 American Major League Soccer
Major League Soccer is the top professional football league in the USA and
Canada. Soccer, as the European football is called in American English, has
always lagged behind other popular sports such as baseball, American football
or basketball. But lately, thanks to the increasing popularity of soccer in the
US, and to the policies that have been implemented, it has lured some of the
![Page 31: An Empirical Investigation of Wage Discrimination in ... · jevili platovou diskriminaci v˚uˇci Afroameriˇcan˚um a hr´aˇc˚um z Latinsk´e Ameriky v nejniˇzˇs´ım decilu](https://reader030.vdocument.in/reader030/viewer/2022040907/5e7d12e667f4fa2fa646aac6/html5/thumbnails/31.jpg)
4. League Overview and Dataset Construction 18
football superstars1. These new acquisitions appear to have very beneficial ef-
fect on the clubs’ financial standing. When we look at the calendar 2016, all
22 teams recorded exponential growth in attendance, viewership, merchandise
sales or social media (Forbes.com 2017). As far as attendance is concerned,
it even exceeds NBA and NHL with average number of 21,692 spectators per
game (Statista.com 2017). MLS’s establishment dates back to the year 1995 as
a condition for United States Soccer Federation to FIFA, which awarded United
States by organizing FIFA World Cup in 1994. Though, the first season was
not played until 1996, with the presence of 10 teams (Mlssoccer.com 1996). In
the first few seasons, the league was not profitable at all and matches were
played on half-empty stadiums. In 2002, two of the founding clubs decided to
cease operations in the league after having financial problems2. The turning
point occured in 2007 when David Beckham, an English superstar and the first
league’s designated player, left Real Madrid to join Los Angeles Galaxy. This
season was also affected by expansion beyond the United States borders. The
first team from Canada that entered the league was Toronto FC. Ever since, the
money pumped into the system has risen dramatically and the league attracted
other teams to participate. As of 2017, the competition consists of 22 teams
and the most successful of them is Los Angeles Galaxy that has won 5 trophies.
As salaries in MLS are limited by a salary cap, clubowners are prohibited to
spend excessive money on compensations. This measure prevents also competi-
tive imbalance among teams. With the introduction of the Designated Player
Rule, though, the wages paid to footballers increased substantially. In 2017,
and for the first time in MLS history, the total guaranteed compensation of
all players exceeded $200 million. The median salary of the whole competition
increased by 15% relative to the season 2016 up to $135,000 and as a result of
the developing salary system, the number of millionaires among MLS players
grew to 28 (Espnfc.com 2017). If we take a look at the table 4.1 which depicts
racial composition of the highest paid players in 2016, we can notice, that two
out of ten players with highest salaries were Hispanics and only one of them
was African American. This ratio appears to be surprising as in the random
1David Beckham started this trend and was subsequently followed by such footballers asSteven Gerrard or Frank Lampard. These two top level British football players transferredfrom English Premier League in 2015 and were signed as designated players. Another famousdesignated player was the Spaniard David Villa who signed in 2014 a contract, that ensuredhim a yearly salary of $5,610,000.
2Both Floridian teams, Tampa Bay Mutiny and Miami Fusion FC, ended their presencein MLS.
![Page 32: An Empirical Investigation of Wage Discrimination in ... · jevili platovou diskriminaci v˚uˇci Afroameriˇcan˚um a hr´aˇc˚um z Latinsk´e Ameriky v nejniˇzˇs´ım decilu](https://reader030.vdocument.in/reader030/viewer/2022040907/5e7d12e667f4fa2fa646aac6/html5/thumbnails/32.jpg)
4. League Overview and Dataset Construction 19
sample that we have collected, black players represent 26.4%.
Table 4.1: The highest paid MLS footballers in 2016
Player Salary Race Position Mins Goals Assists
Ricardo Kaka $7,167,500 Hispanic Midfielder 1955 9 10Sebastian Giovinco $7,115,556 White Striker 2418 17 15Michael Bradley $6,500,000 White Midfielder 2160 1 5Steven Gerrard $6,132,500 White Midfielder 1491 3 11Frank Lampard $6,000,000 White Midfielder 1280 12 3Andrea Pirlo $5,915,690 White Midfielder 2770 1 11David Villa $5,610,000 White Striker 2869 23 4Jozy Altidore $4,825,000 Black Striker 1487 10 5Clint Dempsey $4,605,942 White Midfielder 1429 8 2Giovani dos Santos $4,250,000 Hispanic Striker 2350 14 12
4.2 Salary system
Since the establishment of the league, MLS salaries has been constrained by
a salary cap. Its main purpose was to ensure a competitive balance among
teams. This prevented MLS from attracting players of the highest quality as
the most talented players sought contracts in more generous European clubs.
To ease the strict wage policy and therefore increase quality of players in the
competition, MLS introduced several rules while still sticking to the salary cap.
Designated Player Rule - Also known as Beckham Rule (Reuters 2017) since
David Beckham was the first who used this rule to transfer to Los Angeles
Galaxy. It was adopted in 2007 and it allows MLS teams to sign players
outside their club’s salary cap. The main goal of the policy was to enable
MLS clubs to compete for international players in the football labour
market. Besides higher wages, clubs are also allowed to pay significant
transfer money which would be impossible if there was not for this rule.
Allocation Money - is additional money that is available to football clubs and
therefore each MLS team is entitled to receive this money in addition to
its salary budget. Moreover, for those teams that do not have a third
Designated Player, General Allocation Money (GAM) uses its funds col-
lected by MLS to enable them to purchase more expensive players of
![Page 33: An Empirical Investigation of Wage Discrimination in ... · jevili platovou diskriminaci v˚uˇci Afroameriˇcan˚um a hr´aˇc˚um z Latinsk´e Ameriky v nejniˇzˇs´ım decilu](https://reader030.vdocument.in/reader030/viewer/2022040907/5e7d12e667f4fa2fa646aac6/html5/thumbnails/33.jpg)
4. League Overview and Dataset Construction 20
their desire. In 2017, MLS also introduced the so-called Additional tar-
geted allocation money that allocated additional $1,200,000 for each team
(Mlssoccer.com 2017).
Generation Adidas - This rule was created thanks to the cooperation between
MLS and US soccer. It was aimed to improve the quality of young foot-
ball talent in the US. This programme sponsored by German company
Adidas, covers the salaries of young supertalents who after that do not
have intentions to leave the US. For these players, salary cap does not
count.
The long-term effect of these policies is obvious. Once the restrictions about
salary cap were eased, the trend in terms of wages was mostly upward-sloping.
While in 2006 the sample average compensation of 100 most productive players
was mere $152,845, one decade later the value increased by 401% to $766,481.
The evolution of salaries with top 100 MLS players is depicted in the figures
4.1 and 4.2.
Figure 4.1: MLS average salary of 100 best players
Source: MLS Players Union (2016), author’s computations.
4.3 Dataset Construction
In order to answer our research questions, we have compiled a vast dataset
consisting of 1,100 observations. This large number of observed players will
enable us to focus also on different quantiles of the distribution. From each
![Page 34: An Empirical Investigation of Wage Discrimination in ... · jevili platovou diskriminaci v˚uˇci Afroameriˇcan˚um a hr´aˇc˚um z Latinsk´e Ameriky v nejniˇzˇs´ım decilu](https://reader030.vdocument.in/reader030/viewer/2022040907/5e7d12e667f4fa2fa646aac6/html5/thumbnails/34.jpg)
4. League Overview and Dataset Construction 21
Figure 4.2: Development of players’ median salaries
Source: MLS Players Union (2016), author’s computations.Note: Mind that the seasonal fluctuations might be influenced by the dataset structure.What is important is the long-term trend.
season 2006-2016, we took 100 most productive players in terms of goals and as-
sists because their figures provide reliable information about their performance.
Otherwise, the data generated would not be meaningful. It would be optimal
to include random effect for each individual and arrange the data into a panel.
This method would enable us to account for unique and unobserved character-
istics of each player. Nevertheless, as the majority of the players appear in the
dataset at most twice (see table 4.2), random effects technique would not be
appropriate.
Table 4.2: Occurrence in the dataset
Occurrence 1× 2× 3× 4× 5× 6× 7× 8× 9× 10× 11×
No. of players 382 163 42 18 11 8 4 3 2 1 1
Note: Each number implies the occurence of the players in the sample. Once we sum up therow with the numbers of players, we obtain 635 individuals. If we multiply the number ofplayers by their occurence in the dataset, we get to the 1,100 observations.
Thus, we created a matrix that contains 18 variables out of which 17 are
explanatory. In addition to that, we added 10 dummy variables accounting for
![Page 35: An Empirical Investigation of Wage Discrimination in ... · jevili platovou diskriminaci v˚uˇci Afroameriˇcan˚um a hr´aˇc˚um z Latinsk´e Ameriky v nejniˇzˇs´ım decilu](https://reader030.vdocument.in/reader030/viewer/2022040907/5e7d12e667f4fa2fa646aac6/html5/thumbnails/35.jpg)
4. League Overview and Dataset Construction 22
individual seasons. To examine a possible wage discrimination against non-
white players, we needed to collect three types of data.
Performance data - The first one deals with players’ performance that we need
to control for in order to discover wage inequality given the same produc-
tivity. This data should have the highest effect on player’s compensation
and it is available on the official webpage of MLS. To the best of our
knowledge, no model explaining footballer’s salary has been constructed
and hence, we needed to select the most important variables by intu-
ition. These statistics imply player’s productivity, stamina, activity or
indiscipline.
Salary data - The next type was the information about player’s salary. Amer-
ican sports leagues are unique due to their transparency and they pro-
vide, contrary to the leagues in Europe, information about players’ yearly
salaries. This data is at disposal on the webpage of MLS Players Union.
This association serves as the collective bargaining representative and as
a protection of all players’ rights. For each individual player, they list
his yearly compensation divided into two components - base salary and
guaranteed compensation. The guaranteed compensation is base salary
added to the annualized bonuses and options of players that are included
in the conctract. It is more appropriate for the study as it also includes
the bonuses obtained when signing the contract. Thus, salaries at the end
of each season vary accordingly. Performance bonuses are not included
because there is uncertainty about obtaining these benefits and therefore,
a possible correlation between salaries and performance statistics is ruled
out.
Demographic data - The ethnicity data were observed either from the US sports
channel ESPN or Mlssoccer.com as both of them offer individual pictures
and also information about places of birth as well as the data about
players’ age, which can be used for determining the peak or the bottom
in terms of compensation in their carreers. Since the main question to be
answered is wage discrimination of African Americans and Hispanics, we
were obliged to exclude Asian footballers from the sample3.
We obtained the ultimate dataset that consists of 290 (26.36%) black, 537
(48.82%) white and 273 (24.82%) Latin footballers. These proportions are
3Since there were only 3 Asian players in our dataset, their exlusion could not cause aviolation of random sampling assumption.
![Page 36: An Empirical Investigation of Wage Discrimination in ... · jevili platovou diskriminaci v˚uˇci Afroameriˇcan˚um a hr´aˇc˚um z Latinsk´e Ameriky v nejniˇzˇs´ım decilu](https://reader030.vdocument.in/reader030/viewer/2022040907/5e7d12e667f4fa2fa646aac6/html5/thumbnails/36.jpg)
4. League Overview and Dataset Construction 23
sufficiently high and makes the investigation of any pay gaps possible. While
analysing our data, we found some interesting facts that occured in the seasons
2006-2016:
• Between 2006-2016, 6,175 goals were scored. From that amount, 2817
(45.62%) were recorded by whites, 1727 by blacks (27.97%) and 1631
(26.41%) by Hispanics.
• 9 different players won MLS trophy for the top scorer; only Jeff Cunning-
ham (black) and Chris Wondolowski (white) managed to win this award
twice.
• The best player in terms of productivity was Sebastian Giovinco. Despite
his high guaranteed compensation of $7,115,556 in the season 2015, he
was very contributive for his team Toronto FC as he was able to record
38 points (22 goals and 16 assists).
• In the season 2010, David Beckham played only 466 minutes and scored
2 goals. Given his salary of $6,500,000, one goal cost his club $3,250,000
which is the highest ratio Salary/Goal observed.
• The cheapest goals were scored in the same season by Chris Wondolowski
whose each goal cost only $2,667.
• One minute played of an average black player cost his employer $208.5,
for Hispanic players the amount was $204.2 and for whites $234.1.
• 3 most often fouling players were African Americans; in the season 2013,
Quincy Amarikwa fouled every 17.6 minutes which made him the most
indisciplined player in the league.
![Page 37: An Empirical Investigation of Wage Discrimination in ... · jevili platovou diskriminaci v˚uˇci Afroameriˇcan˚um a hr´aˇc˚um z Latinsk´e Ameriky v nejniˇzˇs´ım decilu](https://reader030.vdocument.in/reader030/viewer/2022040907/5e7d12e667f4fa2fa646aac6/html5/thumbnails/37.jpg)
4. League Overview and Dataset Construction 24
The table 4.3 lists all explanatory variables that were used for the research.
Table 4.3: Definitions of variables
Variable Definition Exp. sign
STRIKER 1 if striker, 0 otherwise ?MIDFIELD 1 if midfielder, 0 otherwise ?GS Games started per season +MINS Minutes played per season +GOALS Goals per season +ASSISTS Assists per season +SHTS Shots per season +SOG Shots on goal per season +FOCO Fouls commited per season -FOSU Fouls suffered per season +YCARD Yellow cards per season -OFFS Offsides per season -STAR 1 if star, 0 otherwise +AGE Age of the player +/-AGE2 Age of the player squared -/+BLACK 1 if black, 0 otherwise ?HISP 1 if Hispanic, 0 otherwise ?YEARt 1 if taken from yeart, 0 otherwise +
The dependent variable is salary in logarithmic form and it is the key ele-
ment for the thesis. Log-level model moderates the effect of outliers emerging
from extreme values in the upper part of the distribution. We expect the com-
pensations to be determined ex post and on that account, we collected salaries
after each season. The table 4.4 represents the distribution of salaries from the
sample.
Table 4.4: Salary - Descriptive statistics
Variable Mean Min 10% 25% 50% 75% 90% Max
Salary ($) 407,315 12,900 48,400 83,562 156,417 258,469 650,000 7,167,500i-th player 1st 110th 275th 550th 825th 990th 1100th
The percentages denote corresponding quantiles
Note: The players are arranged in ascending order.
The reader can notice that the median is significantly lower than the mean.
It is caused by the dramatic difference between 90th and 100th percentile and we
![Page 38: An Empirical Investigation of Wage Discrimination in ... · jevili platovou diskriminaci v˚uˇci Afroameriˇcan˚um a hr´aˇc˚um z Latinsk´e Ameriky v nejniˇzˇs´ım decilu](https://reader030.vdocument.in/reader030/viewer/2022040907/5e7d12e667f4fa2fa646aac6/html5/thumbnails/38.jpg)
4. League Overview and Dataset Construction 25
can deduce that at least 10% of the highest paid players have unproportionally
higher salaries than the rest of the sample. This fact urges us to use quantile
regression in addition to OLS and to estimate the conditional quantiles of the
dependent variable. Moreover, we will be allowed to investigate discrimina-
tion in the lower part of the wage distribution. This approach is reasonable as
discriminating better and more paid players would incur much higher relative
costs for the employer. The table 4.5 offers means of the lower quartile com-
pared to the averages of whole sample. At the first glance we can notice that
while performance statistics do not differ so much, the percentage difference of
salary is immense between these two groups.
Table 4.5: Comparison between whole sample and lower quartile
Salary GS MINS Goals Assists SHTS SOG FoCo FoSu Offs Ycard Age
25th percentile 54,356 16.08 1,456 3.69 2.43 28.38 12.29 22.46 22.85 7.20 2.53 23.94
Whole sample 407,315 19.87 1,765 5.25 3.65 39.04 16.36 25.79 29.18 9.45 2.98 26.55
% difference 649.35% 23.56% 21.19% 42.29% 50.49% 37.56% 33.17% 14.83% 27.72% 31.24% 17.94% 10.89%
To the best of our knowledge, there is no study which would examine the
relation between footballer’s race and wage and hence, the data processing
needs to be done according to studies from another field of sports environment.
For National Basketball League, Brown et al. (1991) introduced the method
of all-career statistics. Their work was extended by Rehnstrom (2009), who
used all-career statistics as well, and found 24.5% premium for white players in
NBA in the season 2008-09. This approach might seem reasonable as employers
are more interested in observing how workers have been doing in a long-term
period. But, professional football is a special case, and in comparison with
basketball, there are many competitions on the top level and statistical com-
parison between them might lead us to false results. The approach of all-career
statistics would not allow us to observe no other players than Americans which
would significantly lower the extent of our work. In addition, some older former
stars towards the end of their career receive relatively lower wages even though
their all-career statistics exceed numbers of other players. For this reason, we
adopted the method of Yang & Lin (2012) who examined the impact of na-
tionality on salaries in NBA. According to them, a player’s current salary is
determined by performance in the previous year. Nevertheless, as compensa-
![Page 39: An Empirical Investigation of Wage Discrimination in ... · jevili platovou diskriminaci v˚uˇci Afroameriˇcan˚um a hr´aˇc˚um z Latinsk´e Ameriky v nejniˇzˇs´ım decilu](https://reader030.vdocument.in/reader030/viewer/2022040907/5e7d12e667f4fa2fa646aac6/html5/thumbnails/39.jpg)
4. League Overview and Dataset Construction 26
tions might not be determined on the basis of one season, our thesis is composed
of two separate regressions - the first examines performance statistics from pre-
vious season and the second deals with average statistics from three previous
seasons. The second technique was used by Hill (2004) in the examination of
racial wage differential in basketball. This method eschews sudden fluctuations
in players’ performance across individual seasons. Furthermore, employers tend
to remunerate players according to their longer consistent performance4.
4.3.1 Performance statistics
After specifying the method of data collection and data transformation, we
can proceed to the description of the data and to the reasoning standing be-
hind its inclusion into the dataset. Goals and assists are likely to be the most
important determinants of the player’s salary since they reflect the most his
importance for the team. Similarly, games started and minutes played per sea-
son imply how much trust the coach has in the player. In addition, the first
mentioned shows player’s readiness and medical condition during the season.
Consequently, we expect all of them to have a positive impact on the salary.
We decided to omit a number of games played because each club is provided
for only 3 substitutions during the game and it is very usual that players are
substituted in the last 15 minutes of the game. The correlation between the
time spent on the pitch and games would therefore decline and it would also
decrease the trustworthiness of the model.
To avoid dummy variable trap, we omitted the position of the defender from
the model5 and this position is therefore included in the intercept. The other
two positions, striker and midfielder, are included as dummy variables and
their effect on wage is difficult to be estimated. The indicators of quality for
players on these two positions differ from indicators for defenders. The other
two variables included in the study are shots and shots on goal per season.
Basically, both of them imply the player’s activity on the pitch and eagerness
for scoring goals. For this reason, we expect them to influence the salary
positively as well. The variable offsides is likely to have a negative impact on
4While compiling the dataset, we noticed, though, that there was no player, whose salarywould remain constant for 3 years and more. Player’s remuneration is therefore often adjustedand changes over time based on previous productivity.
5The position of a goalkeeper was completely excluded from our study as their statisticsare imcomparable with statistics of other players on the pitch.
![Page 40: An Empirical Investigation of Wage Discrimination in ... · jevili platovou diskriminaci v˚uˇci Afroameriˇcan˚um a hr´aˇc˚um z Latinsk´e Ameriky v nejniˇzˇs´ım decilu](https://reader030.vdocument.in/reader030/viewer/2022040907/5e7d12e667f4fa2fa646aac6/html5/thumbnails/40.jpg)
4. League Overview and Dataset Construction 27
the player’s compensation since it rather reflects his inattentiveness. On the
other hand, the variable standing for fouls suffered per season should serve as
a proxy for belligerence and the coefficient is very likely to be positive. If we
move to the position of a defender, there are two statistics that need to be
emphasized. Namely, yellow cards (YCARD) and fouls commited (FOCO).
Both of them indicate indiscipline and their presence should have a negative
effect on the wage.
Figure 4.3: 3-seasons average statistics
Note: We weighted players’ performance based on overall average, which is equal to 1.If the column is higher than 1, it signifies that players of this race hadabove-average performance.
As can be seen from the figure 4.3, both Hispanics and African Americans
earn on average less than their white peers despite their better performance in
terms of goals and shots. From the table 4.6, which captures statistics based
on 1-season observation, we can spot that the salary of whites and blacks in the
middle of the distribution is almost the same, though. Median salaries of both
are significantly exceeded by salaries of Hispanics. Both blacks and Hispanics
appear to be more undisciplined which can be seen in higher number of offsides
as well as fouls commited. Blacks were also given 7.4% less time on the pitch
than whites. However, based on these results, we cannot conclude anything
about discriminatory behaviour. If we want to detect any premium for play-
ers of different ethnicity, we need to control for the player’s performance, and
![Page 41: An Empirical Investigation of Wage Discrimination in ... · jevili platovou diskriminaci v˚uˇci Afroameriˇcan˚um a hr´aˇc˚um z Latinsk´e Ameriky v nejniˇzˇs´ım decilu](https://reader030.vdocument.in/reader030/viewer/2022040907/5e7d12e667f4fa2fa646aac6/html5/thumbnails/41.jpg)
4. League Overview and Dataset Construction 28
therefore descriptive statistics are not sufficient.
Table 4.6: Sample means, medians & standard deviations - 1 season
1-season statistics Black White Hispanics OverallVariable Mean Median Mean Median Mean Median Mean Median
Salary ($) 367,631 149,667 434,152 150,000 377,723 175,000 407,315 156,417(212229) (1123149) (719240) (974237)
Games started 19.73 20 21.36 23 21.15 23 20.89 22(7.68) (7.52) (6.75) (7.41)
Minutes 1,763 1,796 1,894 1,980 1,850 1,955 1,849 1,924(640.77) (638.71) (565.19) (624.24)
Goals 5.96 5 5.21 4 5.97 5 5.61 4(4.27) (3.89) (3.60) (3.94)
Assists 3.21 3 3.78 3 4.38 4 3.78 3(2.57) (3.31) (3.42) (3.19)
Shots 41.51 36 38.79 34 44.32 42 40.99 37(25.71) (24.54) (22.44) (24.45)
Shots on goal 17.49 15 16.16 14 18.3 17 17.09 15(10.87) (10.93) (9.75) (10.67)
Fouls commited 26.73 25 25.42 24 28.26 27 26.49 25(13.59) (12.22) (13.11) (12.87)
Fouls suffered 30.32 28 27.86 25 33.38 29 29.88 27(16.96) (14.90) (18.51) (16.56)
Offsides 11.99 10 8.09 5 10.41 7 9.72 6(10.47) (9.99) (10.00) (9.99)
Yellow cards 2.79 2 3.13 3 3.43 3 3.12 3(2.08) (2.28) (2.08) (2.19)
Age 25.73 25 26.65 26 27.18 27 26.55 26(4.17) (4.16) (4.71) (4.34)
N 290 537 273 1100
standard deviations are noted in parentheses
![Page 42: An Empirical Investigation of Wage Discrimination in ... · jevili platovou diskriminaci v˚uˇci Afroameriˇcan˚um a hr´aˇc˚um z Latinsk´e Ameriky v nejniˇzˇs´ım decilu](https://reader030.vdocument.in/reader030/viewer/2022040907/5e7d12e667f4fa2fa646aac6/html5/thumbnails/42.jpg)
4. League Overview and Dataset Construction 29
Table 4.7 illustrates descriptive statistics where players’ performance was
observed and averaged for the 3 previous seasons. As we can see from the lower
standard deviations, this treatment serves as a robustness check. Compared to
the table 4.6, most of the values are slighly lower. Otherwise, both approaches
appear to have very similar descriptive statistics. Players of black skin played
on average 138 minutes less than whites and 125 minutes less than players
from Latin America. Hispanics were more active than the rest of the dataset
as their average player scored 19.88% more goals than his white counterpart.
They also recorded better results concerning assists, shots or shots on goal.
On the contrary, Southerner temperament is manifested in terms of commited
fouls and yellow cards where they outdid other footballers.
Table 4.7: Sample means, medians & standard deviations - 3 seasons
3-seasons statistics Black White Hispanics OverallVariable Mean Median Mean Median Mean Median Mean Median
Salary ($) 367,631 149,667 434,152 150,000 377,723 175,000 407,315 156,417(212229) (1123149) (719240) (974237)
Games started 18.55 19.33 20.28 21.33 20.49 21.50 19.87 20.83(6.79) (6.74) (6.05) (6.64)
Minutes 1,666 1,711 1,804 1,873 1,791 1,879 1,765 1,824(567.53) (571.08) (512.17) (559.24)
Goals 5.54 4.33 4.83 4 5.79 5 5.25 4.33(3.77) (3.46) (3.45) (3.57)
Assists 2.99 2.67 3.65 3 4.36 4 3.65 3(2.04) (2.76) (3.04) (2.71)
Shots 39.10 34.75 37.18 33.67 42.66 40 39.04 36(22.99) (22.30) (20.60) (22.19)
Shots on goal 16.46 15 15.57 14 17.82 16.67 16.36 14.83(9.52) (9.98) (9.29) (9.73)
Fouls commited 25.77 24.25 24.88 23.50 27.60 26.67 25.79 24.58(11.89) (10.68) (11.74) (11.33)
Fouls suffered 28.95 28 27.34 25 33.07 29 29.18 26.50(1.68) (1.90) (1.84) (1.85)
Offsides 11.17 10 8.10 5 10.28 7 9.45 6.67(9.02) (8.73) (9.56) (9.12)
Yellow cards 2.65 2.50 2.94 3 3.39 3 2.98 3(1.68) (1.90) (1.84) (1.85)
Age 25.73 25 26.65 26 27.18 27 26.55 26(4.17) (4.16) (4.71) (4.34)
N 290 537 273 1100
standard deviations are noted in parentheses
4.3.2 Other Factors
Undoubtedly, performance statistics are relevant in determining player’s wage,
but there are other factors to be considered. Age is one of them and it def-
![Page 43: An Empirical Investigation of Wage Discrimination in ... · jevili platovou diskriminaci v˚uˇci Afroameriˇcan˚um a hr´aˇc˚um z Latinsk´e Ameriky v nejniˇzˇs´ım decilu](https://reader030.vdocument.in/reader030/viewer/2022040907/5e7d12e667f4fa2fa646aac6/html5/thumbnails/43.jpg)
4. League Overview and Dataset Construction 30
initely should not be missed out. With increasing age, players should obtain
more experience and be rewarded by higher wages. This effect must be, though,
accompanied with a decreasing marginal effect. At one point of a career, play-
ers tend to lose their physical strength and stamina, which leads to an overall
decrease in performace. Holmes (2011) uncovered an increasing effect of age on
the salary of MLB players. The turning point in his study occurred at the age
of 17.6 years, the estimates turned out to be statistically insignificant, though.
On the other hand, Yang & Lin (2012) found a decreasing effect of age on the
salary in American NBA. As another covariate in their salary equation, they
included the variable star that was equal to 1 if the player was chosen into
NBA All-Star in the previous season. We adopted the same approach since
this variable control for a possible fact that football superstars might be remu-
nerated more than appropriate for their performance on the pitch.
The last set of variables concerns the years from which the data were taken.
These dummy variables include besides other things the economic situation in
the US as well as the popularity of soccer there. As the statistics were collected
in seasons 2006-2016, our model contains ten dummies and the base year 2006
is therefore included in the intercept. These variables are expected to be of a
high magnitude and significance for the reasons stated earlier.
![Page 44: An Empirical Investigation of Wage Discrimination in ... · jevili platovou diskriminaci v˚uˇci Afroameriˇcan˚um a hr´aˇc˚um z Latinsk´e Ameriky v nejniˇzˇs´ım decilu](https://reader030.vdocument.in/reader030/viewer/2022040907/5e7d12e667f4fa2fa646aac6/html5/thumbnails/44.jpg)
Chapter 5
Model
In the first part of this chapter we justify the aproach that has been chosen.
After justifying the methodologies, we will move on to present our model that
estimates the determinants of a player’s salary.
5.1 General Model
To answer the research questions, we have collected a vast set of data from
several seasons. To be able to estimate the development of salaries in football
enviroment, we monitor data of 1,100 different observations during eleven sea-
sons, which means that we face a combination of cross-sectional and time series
data. According to Wooldridge (2012), the data with cross-sectional and time
series aspects can be arranged in two kinds of data sets. If we obtain informa-
tion from a random sample at different points in time, this data is called an
independently pooled cross section. On the other hand, collecting data from
the same individuals leads to a panel dataset. The assumption of observing
same units in time is not satisfied in our data and hence, we arrange it into
independently pooled cross section.
log(Salary) = α +X~β +R~γ +T~δ + ~u (5.1)
In this thesis, we utilize the explained variable (salary) in a logarithmic
functional form as we want to know the percentage change in salary. At the
same time, once we utilize the logarithm transformation, we take into consid-
eration the heteroskedasticity that arises from the large variance in salaries.
Using log-level model means that the interpretation must be changed and each
coefficient βi, γi and δi multiplied by 100, describes a percentage increase or
![Page 45: An Empirical Investigation of Wage Discrimination in ... · jevili platovou diskriminaci v˚uˇci Afroameriˇcan˚um a hr´aˇc˚um z Latinsk´e Ameriky v nejniˇzˇs´ım decilu](https://reader030.vdocument.in/reader030/viewer/2022040907/5e7d12e667f4fa2fa646aac6/html5/thumbnails/45.jpg)
5. Model 32
decrease in player’s wage. Vector X is a vector of size k × n as there are k
explanatory variables standing for performance, position or age, and n obser-
vations. R is a vector implying the race of each individual player and finally T
is a vector of time dummies which is equal to 1 if observed in the particular year.
Even though the OLS method was used by the majority of researches deal-
ing with racial discrimination in sport (Kahn 1991), we test the satisfaction of
the assumptions in appendices and also point to possible problems.
Before we move to our model, there is one issue that needs to be considered
- wage is not in real, but in nominal dollars. As wages in our dataset increase
also due to inflation, we should adjust them in order to examine solely the
effect on real wages. This procedure requires deflating wages from 2007-2016
to 2006 dollars. Fortunately, this turns out to be unnecessary as far as we use
log-linear model. The difference between real and nominal wage can only be
seen in the form of different coefficients on the season dummies. We provide
an explanation with a simplified model using only two seasons.
log(wagei/PS2) = log(wagei)− log(PS2) (5.2)
Where the constant PS2 describes the price level change in the season 2
relative to the season 1. While wagei varies across players, PS2 remains the
same and therefore, log(PS2) is basically absorbed into the intercept. The
bottom line is that, nominal wages do not have to be turned into real ones for
studying the determinants of footballers’ compensation.
5.2 Our Model
As we noted earlier, some of the research questions about pay discrimination
will be answered using Ordinary Least Squares method. Our model has follow-
ing form:
![Page 46: An Empirical Investigation of Wage Discrimination in ... · jevili platovou diskriminaci v˚uˇci Afroameriˇcan˚um a hr´aˇc˚um z Latinsk´e Ameriky v nejniˇzˇs´ım decilu](https://reader030.vdocument.in/reader030/viewer/2022040907/5e7d12e667f4fa2fa646aac6/html5/thumbnails/46.jpg)
5. Model 33
log(Salary) =β0 + β1Goals+ β2Assists+ β3SoG+ β4Shots+ β5Fouls+
β6Y Card+ β7Offs+ β8Mins+ β9GS + β10FoCo+ β11Age+
β12Age2 + β13Striker + β14Midfield+ γ1Black + γ2Hisp+
δ0year2007 + δ1year2008 + δ2year2009 + δ3year2010+
δ4year2011 + δ5year2012 + δ6year2013 + δ7year2014+
δ8year2015 + δ9year2016 + uit
(5.3)
The dependent variable log(Salary) denotes logarithm of individual salaries.
This form was chosen since we are interested in percentage rather than abso-
lute change in player’s wage. For instance, one more goal scored increases or
decreases salary by β1100%. The error term uit stands for speciffic, unobserved
effect of each player.
The main concern of this study is expressed by dummy variables Black
and Hisp. If these two variables turn out to be significant, this study will dis-
close either positive or negative pay discrimination relative to white players. A
positive coefficient would therefore imply pay premium either for black or His-
panic players. In case of negative coefficient, we will speak about racial wage
discrimination against non-whites. Besides racial dummies, we also included
dummy variables that control for individual years. As we mentioned earlier,
these variable contain information about time trends. Omitting them would
cause a serious problem for our model, since for example, the value of dollar
in 2006 is not equal to its value in 2016. Presumably, years will also be of sta-
tistical significance. The increasing importance and popularity of professional
football in the latest years attracted many investors who pumped money into
the system.
![Page 47: An Empirical Investigation of Wage Discrimination in ... · jevili platovou diskriminaci v˚uˇci Afroameriˇcan˚um a hr´aˇc˚um z Latinsk´e Ameriky v nejniˇzˇs´ım decilu](https://reader030.vdocument.in/reader030/viewer/2022040907/5e7d12e667f4fa2fa646aac6/html5/thumbnails/47.jpg)
Chapter 6
Results
As was said earlier, we used two different regression analyses1 to detect racial
discrimination in professional football. The most important lesson is that the
MLS clubs turned out to have discriminatory behaviour. The table 6.4 sum-
marizes wage gaps at the individual parts of the salary distribution and the
estimates indicate that Latinos and African Americans are deprived of a sig-
nificant part of their salary due to their origins. Tables 6.3 and 6.2 repre-
sent the results of OLS estimates. For both methods of data transformation,
we run 4 individual regressions2 to answer the research question we posed in
the thesis proposal. The results from both approaches are slightly different
due to a different nature of the data. Since we detected heteroskedacity and
OLS assumption of homoskedasticity was violated (Appendix A), we obtained
also heteroskedasticity-robust standard errors (see table B.1 in Appendix B)
to show that most of the variables have very similar t-statistics and statisti-
cal significance. To account for non-normality of error terms, we also ran a
robust regression that dampens the impact of excessively high values in the
upper part of the salary distribution. The results are introduced in Appendix
B and demonstrate that the inference about racial discrimination remained un-
changed. We will now focus on the description of OLS results, though, since it
is the common method in the existing literature when examining pay discrim-
ination between races.
1OLS and quantile regressions.2The Model 1 is the core regression that helps us to answer the question about discrim-
ination of the average player. Models 2 and 3 deal with the effect of increasing wage levelon the pay gap and finally model 4 clarifies whether or not the players are discriminated ondifferent positions.
![Page 48: An Empirical Investigation of Wage Discrimination in ... · jevili platovou diskriminaci v˚uˇci Afroameriˇcan˚um a hr´aˇc˚um z Latinsk´e Ameriky v nejniˇzˇs´ım decilu](https://reader030.vdocument.in/reader030/viewer/2022040907/5e7d12e667f4fa2fa646aac6/html5/thumbnails/48.jpg)
6. Results 35
6.1 Discussion of the results
As the principle of 3 seasons has slightly higher adjusted R2 and therefore,
the variance in salary is better explained by the explanatory variables, we will
be describing results of the table 6.2 in the following text. Dummy variables
BLACK and HISP have both negative coefficient once we examine the av-
erage player in the horizon of three seasons. However, this is not where we
should look for wage discrimination because better and more competitive play-
ers would easily transfer to another club if they felt like being discriminated
against. Hence, the coefficients for racial dummies are not statistically signifi-
cant and we would not be able to reject the null hypothesis H0 : γ1 = γ2 = 0.
As a very interesting outcome we can consider coefficients for positions.
Given the same performance, midfielders earn on average 16.1 % less than
defenders at 5% significance level. This benefit might emerge from the fact
that we could not control for other variables that reflect quality of defenders.
Such factors could be for example a number of tackles or a number of won aerial
duels. Unfortunately, this data is not provided and it is inaccessible for our
study. Model 4 takes into account possible interactions between the player’s
race and position and it consists of four multiplications of two dummy variables.
Their presence should account for different physical aspects of each race. As we
see in the table 6.1, physically stronger African Americans are more likely to be
centre forwards whereas relatively weeker Latinos are more often to be found
as wingers or centre midfielders. The effect of these interactions is that for each
Table 6.1: Relationship between race and position
Race/Position Defender Midfielder Striker
Black 12.46% 29.75% 57.79%White 14.37% 44.96% 40.67%Hispanic 5.88% 44.86% 49.26%
position and race the intercept differs. For example, for a midfielder from Latin
America, the intercept would be: 12.126− 0.164 + 0.160− 0.322 = 11.800.
The reader can notice that none of the interactions is significicant3 and the value
3As the assumptions of normality is violated and t-statistics using OLS are no longer valid,we checked the significance using a robust regression. The inference about insignificanceremained the same.
![Page 49: An Empirical Investigation of Wage Discrimination in ... · jevili platovou diskriminaci v˚uˇci Afroameriˇcan˚um a hr´aˇc˚um z Latinsk´e Ameriky v nejniˇzˇs´ım decilu](https://reader030.vdocument.in/reader030/viewer/2022040907/5e7d12e667f4fa2fa646aac6/html5/thumbnails/49.jpg)
6. Results 36
of adjusted R2 indicates that the explained variance of the players’ income
remained the same after adding these four predictors into the model. Not
surprisingly, the more games a player is included in the starting line-up, the
more money he is rewarded. The high correlation of 83.28% between games
and minutes results in a very surprising outcome.
Figure 6.1: Correlation between GS and MINS
0 5 10 15 20 25 30
50
01
00
02
00
03
00
0
Games in the starting line−up
Min
ute
s p
laye
d p
er
se
aso
n
Table 6.2 shows, that if all other predictors are kept constant, minutes
played have a negative impact on wage. In case of real-life football, other ex-
planatory variables tend to change as well when the predictor minutes changes.
Consequently, estimating changes just for this variable might be unnatural and
we need to be very careful about the interpretation. A false inference would be
to assume that players who play less, earn more money. The coefficient rather
says that players who manage to perform the same work during less time, and
are more time-efficient, are likely to obtain higher compensation.
Goals and assists were both found to be positive and highly statistically sig-
nificant. This is not surprising because these two variables imply the player’s
contribution for the team. Similarly, ten more shots per season are supposed to
increase salary by 7%. On the contrary, despite the positive sign, shots on goal
are insignificant for the model. Adverse effect was expected from the variables
fouls commited and offsides as they both reflect indiscipline and incaution. In
the case of fouls, we would even be able to reject the null hypothesis at 1%
![Page 50: An Empirical Investigation of Wage Discrimination in ... · jevili platovou diskriminaci v˚uˇci Afroameriˇcan˚um a hr´aˇc˚um z Latinsk´e Ameriky v nejniˇzˇs´ım decilu](https://reader030.vdocument.in/reader030/viewer/2022040907/5e7d12e667f4fa2fa646aac6/html5/thumbnails/50.jpg)
6. Results 37
level. The impact of fouls suffered is rather negligible as being fouled once more
increases the salary only by 0.4%. However, since this coefficient is significant,
we can deduce that clubs appreciate combative and dribbling players who do
not hesitate to get their teams into a dangerous situation at the cost of being
fouled. Surprisingly, a positive sign was also obtained for yellow cards. The
premium of 6.82% is not directly related to the amount of yellow cards, but
this variable might correlate with the player’s activity. As far as star players
are concerned, we came to the similar conclusion as Yang & Lin (2012) who
also found large magnitude and significance in American NBA.
The estimated coefficients imply that age has an increasing effect on wage.
To find the turning point, we need to use first derivative to find an extreme:
0 = β1 + 2β2age ⇒ age∗ = −(β1/2β2)
.
After filling in values from our model, we get age∗ = − −0.159682×0.00461
= 17.32
years which is considered to be the lowest bottom of player’s career. Pre-
dictable outcomes were received for dummy variables that controlled for time
trend. Since 2010, the increasing popularity of MLS jointly with new salary
policies has caused year dummies to have a very large magnitude and signif-
icance. Nevertheless, increased money in the system did not cause any wage
gap across races, which can be seen from models 2 and 3. We used the DID
estimator to study the differential effect between the treatment group (non-
whites) and the control group (whites) and how this effect changes for the
average player after the salary cap easing. As a system shock, we considered
the season 2010 because this year appears to be the break-even point for the
MLS in terms of money spent on wages. Even though the level of wages raised
substantially and brought an average premium of 34.58%, the easing of salary
cap turned out to be non-discriminatory. Nevertheless, if we wanted to ex-
amine the establishment of the Designated Player Rule before and after this
policy was introduced, we would have to interact the season 2007. This year,
though, does not seem to have any influence on the guaranteed compensation
of an average player. Furthermore, we only controlled for one season before
this adjustment of salary system was presented. If one would like to examine
the exact effect of Beckham’s rule on the wage gap, one season is not enough.
![Page 51: An Empirical Investigation of Wage Discrimination in ... · jevili platovou diskriminaci v˚uˇci Afroameriˇcan˚um a hr´aˇc˚um z Latinsk´e Ameriky v nejniˇzˇs´ım decilu](https://reader030.vdocument.in/reader030/viewer/2022040907/5e7d12e667f4fa2fa646aac6/html5/thumbnails/51.jpg)
6. Results 38
Unfortunately, we were limited by the data unavailability for salaries which are
not listed before 2006.
In total, there are 7 explanatory variables4 that are not significant for the
equation. To check for their joint significance, we ran a F-test, where
F = [(0.5319− 0.5297)/(1− 0.5319)]× (1, 072/7) = 0.7179
This result is below the 5% critical value and we cannot reject the null hypoth-
esis of statistical insignificance. In other words, these variables are redundant
for the model and the appropriate technical approach would be to exclude
them from our research. However, their presence is crucial as racial dummies
are of the main interest for the thesis and without them, our hypothesis would
lose its meaning. Furthermore, season dummy variables include inflation and
other both economic and non-economic factors. Their exclusion would lead to
a comparison of nominal instead of real values of dollar. We provide at least
the table C.1 (see Appendix C) that compares results between the restricted
and unrestricted models. The restricted model could be used, if we were not
interested in investigating the wage differential between races, but only in the
determinants of the player’s salary. We can see that the changes in estimates
are negligible and including insignificant predictors does not cause any larger
biases. Consequently, remaining two covariates were not omitted as they pro-
vide a valuable information about player’s position, activity and accuracy.
As we move to the 1-season statistics OLS results, some changes were
recorded as the adjusted R2 dropped by two percentage points. While the
majority of predictors have slightly low t-statistics, the variable implying how
many shots a player had per season gained on its significance. On the other
hand, the variable midfielder was no longer found to be significant at 10%
significance level and its magnitude also dropped compared to the 3-seasons
records. Since other estimates, including racial dummies, remained very simi-
lar, we proceed to the outcome of quantile regression which offers much more
interesting discoveries.
4Striker, Shots on goal, Black, Hispanic, year 2009, year 2008 and year 2007.
![Page 52: An Empirical Investigation of Wage Discrimination in ... · jevili platovou diskriminaci v˚uˇci Afroameriˇcan˚um a hr´aˇc˚um z Latinsk´e Ameriky v nejniˇzˇs´ım decilu](https://reader030.vdocument.in/reader030/viewer/2022040907/5e7d12e667f4fa2fa646aac6/html5/thumbnails/52.jpg)
6. Results 39
Table 6.2: OLS regressions of 3-seasons statistics
Model 1 Model 2 Model 3 Model 4Variable Estimate (t-stats) Estimate (t-stats) Estimate (t-stats) Estimate (t-stats)
Constant 12.127*** (15.240) 12.125*** (15.229) 12.140*** (15.226) 12.126*** (15.166)STRIKER -0.152 (-1.545) -0.152 (-1.545) -0.153 (-1.549) -0.204* (-1.678)MIDFIELD -0.176** (-2.053) -0.176** (-2.053) -0.176** (-2.057) -0.164 (-1.536)GS 0.063*** (2.653) 0.063*** (2.656) 0.063*** (2.665) 0.063*** (2.697)MINS -0.001*** (-3.141) -0.001*** (-3.144) -0.001*** (-3.152) -0.001*** (-3.245)GOALS 0.059*** (3.733) 0.058*** (3.734) 0.059*** (3.736) 0.057*** (3.648)ASSISTS 0.045*** (3.975) 0.045*** (3.969) 0.045*** (3.953) 0.048*** (4.162)SHTS 0.007* (1.850) 0.007* (1.835) 0.007* (1.820) 0.007* (1.842)SOG 0.005 (0.459) 0.005 (0.466) 0.005 (0.478) 0.006 (0.572)FOCO -0.008*** (-2.691) -0.008*** (-2.686) 0.008*** (-2.672) -0.008*** (-2.705)FOSU 0.004* (1.905) 0.004* (1.907) 0.004* (1.910) 0.004** (1.965)YCARD 0.066*** (3.842) 0.066*** (3.839) 0.066*** (3.840) 0.067*** (3.909)OFFS -0.009** (-2.150) -0.009*** (-2.154) -0.009** (-2.150) -0.009** (-2.259)STAR 0.571*** (8.226) 0.571*** (8.224) 0.571*** (8.224) 0.561*** (8.000)AGE -0.160*** (-2.687) -0.160*** (-2.683) -0.161*** (-2.698) -0.158*** (-2.663)AGE2 0.005*** (4.241) 0.005*** (4.236) 0.005*** (4.249) 0.004*** (4.221)BLACK -0.024 (-0.411) -0.021 (-0.350) -0.024 (-0.409) -0.137 (-0.885)HISP -0.055 (-0.905) -0.055 (-0.908) -0.060 (-0.950) 0.160 (0.774)year2016 0.774*** (6.677) 0.782*** (6.119) 0.761*** (6.126) 0.788*** (6.775)year2015 0.854*** (7.408) 0.854*** (7.404) 0.856*** (7.410) 0.865*** (7.471)year2014 0.581*** (5.015) 0.581*** (5.014) 0.583*** (5.021) 0.596*** (5.117)year2013 0.551*** (4.785) 0.551*** (4.784) 0.553*** (4.791) 0.564*** (4.875)year2012 0.549*** (4.888) 0.549*** (4.887) 0.550*** (4.893) 0.564*** (5.001)year2011 0.366*** (3.279) 0.366*** (3.280) 0.367*** (3.286) 0.378*** (3.375)year2010 0.297*** (2.689) 0.297*** (2.686) 0.297*** (2.694) 0.310*** (2.805)year2009 0.136 (1.238) 0.136 (1.237) 0.136 (1.244) 0.143 (1.306)year2008 0.035 (0.322) 0.035 (0.321) 0.035 (0.324) 0.039 (0.364)year2007 0.001 (0.005) 0.001 (0.005) 0.001 (0.007) 0.003 (0.031)BLACK*year2010 - 0.220 (1.256) - -HISP*year2010 - - -0.214 (-1.109) -BLACK*STRIKER - - - 0.162 (0.931)BLACK*MIDFIELD - - - 0.101 (0.561)HISP*STRIKER - - - -0.149 (-0.663)HISP*MIDFIELD - - - -0.322 (-1.441)
R2 0.532 0.532 0.532 0.534Adjusted R2 0.520 0.520 0.520 0.520N 1100 1100 1100 1100F-statistic (DF) 45.11 (27; 1072) 43.46 (28; 1071) 43.47 (28; 1071) 39.45 (31; 1068)
*Significant at 10% level, **Significant at 5% level, ***Significant at 1% level
![Page 53: An Empirical Investigation of Wage Discrimination in ... · jevili platovou diskriminaci v˚uˇci Afroameriˇcan˚um a hr´aˇc˚um z Latinsk´e Ameriky v nejniˇzˇs´ım decilu](https://reader030.vdocument.in/reader030/viewer/2022040907/5e7d12e667f4fa2fa646aac6/html5/thumbnails/53.jpg)
6. Results 40
Table 6.3: OLS regressions of 1-season statistics
Model 1 Model 2 Model 3 Model 4Variable Estimate (t-stats) Estimate (t-stats) Estimate (t-stats) Estimate (t-stats)
Constant 12.255*** (15.109) 12.255*** (15.100) 12.252*** (15.077) 12.236*** (15.011)STRIKER -0.087 (-0.891) -0.087 (-0.891) -0.087 (-0.889) -0.126 (-1.041)MIDFIELD -0.125 (-1.473) -0.126 (-1.472) -0.125 (-1.469) -0.105 (-0.982)GS 0.035* (1.789) 0.035* (1.786) 0.035* (1.781) 0.035* (1.793)MINS -0.001** (-2.319) -0.001** (-2.314) -0.001** (-2.308) -0.001** (-2.374)GOALS 0.043*** (3.546) 0.043*** (3.544) 0.043*** (3.545) 0.043*** (3.508)ASSISTS 0.032*** (3.609) 0.032*** (3.606) 0.032*** (3.608) 0.033*** (3.712)SHTS 0.009*** (3.018) 0.009*** (3.012) 0.009*** (3.016) 0.009*** (3.054)SOG -0.007 (-0.812) -0.007 (-0.811) -0.007 (-0.814) 0.006 (-0.760)FOCO -0.006** (-2.168) -0.006** (-2.166) -0.006** (-2.168) -0.006*** (-2.239)FOSU 0.006*** (2.934) 0.006*** (2.933) 0.006*** (2.931) 0.006*** (2.992)YCARD 0.028** (2.049) 0.028*** (2.048) 0.028** (2.049) 0.029** (2.142)OFFS -0.005 (-1.509) -0.005 (-1.508) -0.005 (-1.510) -0.006 (-1.601)STAR 0.730*** (9.951) 0.730*** (9.945) 0.730*** (9.945) 0.724*** (9.800)AGE -0.169*** (-2.803) -0.169*** (-2.801) -0.169*** (-2.793) -0.167*** (-2.763)AGE2 0.005*** (4.413) 0.005*** (4.410) 0.005*** (4.398) 0.005*** (4.377)BLACK -0.045 (-0.766) -0.045 (-0.733) -0.045 (-0.766) -0.122 (-0.778)HISP -0.029 (-0.475) -0.029 (-0.475) -0.027 (-0.429) 0.145 (0.692)year2016 0.770*** (6.570) 0.770*** (5.955) 0.773*** (6.157) 0.778*** (6.625)year2015 0.887*** (7.730) 0.887*** (7.727) 0.887*** (7.717) 0.893 (7.759)year2014 0.557*** (4.814) 0.557*** (4.812) 0.556*** (4.803) 0.565*** (4.869)year2013 0.495*** (4.283) 0.495*** (4.784) 0.494*** (4.275) 0.502*** (4.335)year2012 0.499*** (4.400) 0.549*** (4.281) 0.499*** (4.392) 0.508*** (4.468)year2011 0.302*** (2.652) 0.302*** (2.650) 0.302*** (2.646) 0.309*** (2.708)year2010 0.287** (2.544) 0.286** (2.543) 0.287** (2.540) 0.296*** (2.615)year2009 0.168 (1.499) 0.168 (1.498) 0.167 (1.495) 0.173 (1.542)year2008 0.021 (0.187) 0.021 (0.187) 0.021 (0.186) 0.023 (0.206)year2007 0.062 (0.563) 0.062 (0.563) 0.062 (0.562) 0.061 (0.557)BLACK*year2010 - 0.220 (1.256) - -HISP*year2010 - - -0.214 (-1.109) -BLACK*STRIKER - - - 0.122 (0.687)BLACK*MIDFIELD - - - 0.056 (0.302)HISP*STRIKER - - - -0.113 (-0.497)HISP*MIDFIELD - - - -0.265 (-1.168)
R2 0.514 0.514 0.514 0.515Adjusted R2 0.502 0.501 0.501 0.501N 1100 1100 1100 1100F-statistic (DF) 41.98 (27; 1072) 40.44 (28; 1071) 40.44 (28; 1071) 36.61 (31; 1068)
*Significant at 10% level, **Significant at 5% level, ***Significant at 1% level
![Page 54: An Empirical Investigation of Wage Discrimination in ... · jevili platovou diskriminaci v˚uˇci Afroameriˇcan˚um a hr´aˇc˚um z Latinsk´e Ameriky v nejniˇzˇs´ım decilu](https://reader030.vdocument.in/reader030/viewer/2022040907/5e7d12e667f4fa2fa646aac6/html5/thumbnails/54.jpg)
6. Results 41
Even though the race has not played an important role for the OLS equa-
tion, we cannot speak about non-discriminatory behaviour in MLS because the
problem of wage differentials might be rather rooted in the bottom part of the
salary distribution as was uncovered by Holmes (2011) in Major League Base-
ball. He argued that premia for whites in the lower subset of the population
is caused by relatively small costs for the employer because the performance of
these players is not likely to be important for the team. Consequently, adopting
the same method, we uncovered that salary differences for the players in the
bottom decile and quartile are much higher and they are significant at the sig-
nificance level where α = 0.05. The final pay premia for whites were calculated
using the transformation 100(exp(γ1,2) − 1)% that gives us the exact percent-
age change in predicted salaries (Wooldridge 2015). Once players’ performance
was observed for three seasons, blacks in the bottom decile received 18.86% less
than their white counterparts. We can see that this difference falls as we get
to the upper part of the salary distribution and for τ = 0.9, there is even pay
premium equal to 8.87% in favour of black players. However, the coefficients
lose statistical significance once we exceed τ = 0.25. A similar situation occur
as far as Hispanic players are concerned. While the poorest 10% of them are
deprived of 15.3% from their salaries compared to whites, the wage differential
for the median player is only 6.85% and not statistically significant. This coef-
ficient even goes to positive values once we examine the upper quartile.
Table 6.4: Results of Quantile Regression
1-season statistics 3-seasons statistics
Black Hispanics Black HispanicsQuantile Estimate Wage gap Estimate Wage gap Estimate Wage gap Estimate Wage gap
10% -0.090* -8.61% -0.142** -13.24% -0.209*** -18.86% -0.166*** -15.30%(-1.953) (-2.422) (-4.792) (-3.021)
25% -0.078* -7.50% -0.010 -1.00% -0.120** -11.31% -0.101** -9.61%(-1.727) (-0.179) (-2.570) (-2.214)
50% -0.023 -2.27% 0.086 8.98% -0.031 -3.05% -0.071 -6.85%(-0.360) (1.322) (-0.547) (-1.090)
75% -0.023 -2.27% 0.043 4.39% 0.008 0.80% 0.019 1.92%(-0.356) ( 0.527) (0.103) (0.253)
90% 0.056 5.76% 0.041 4.19% 0.085 8.87% 0.071 7.36%(0.643) (0.546) (0.915) (1.063)
t-statistics are noted in parentheses
*Significant at 10% level, **Significant at 5% level, ***Significant at 1% level
![Page 55: An Empirical Investigation of Wage Discrimination in ... · jevili platovou diskriminaci v˚uˇci Afroameriˇcan˚um a hr´aˇc˚um z Latinsk´e Ameriky v nejniˇzˇs´ım decilu](https://reader030.vdocument.in/reader030/viewer/2022040907/5e7d12e667f4fa2fa646aac6/html5/thumbnails/55.jpg)
6. Results 42
Figure 6.2: Quantile regression coefficients
Note: The axis x describes the corresponding quantile while the axis y shows the magnitudeof coefficient. The grey area represents a 90% confidence interval (CI), so if the interval doesnot include zero, the coefficient is in a given quantile significantly different from zero. Theline is from OLS and the dashed lines represent 90% CI. All graphs perfectly demonstratethat the coefficients change as we examine different parts of the distribution.
![Page 56: An Empirical Investigation of Wage Discrimination in ... · jevili platovou diskriminaci v˚uˇci Afroameriˇcan˚um a hr´aˇc˚um z Latinsk´e Ameriky v nejniˇzˇs´ım decilu](https://reader030.vdocument.in/reader030/viewer/2022040907/5e7d12e667f4fa2fa646aac6/html5/thumbnails/56.jpg)
Chapter 7
Conclusion
This bachelor thesis investigated wage discrimination in professional football
using two different econometric methods. The first one is called Ordinary Least
Squares and it estimates the effect of race for an average player in the popula-
tion. On the other hand, quantile regression offers different and more complex
data analysis as it enables to examine the effect for individual points of the
salary distribution. For both methods, we used the same variables about play-
ers’ performance, age, race and also the season from which given salary was
observed. Salaries were observed ex post and thanks to the explanatory vari-
ables included into our model, we were able to explain more than 50% of the
variance in the dependent variable.
To investigate a potential racial salary discrimination, we have collected
seasonal MLS data from 2004-2016. Our large dataset, composed of a rich
set of explanatary variables, was focused on the most productive players from
individual seasons. We discovered a strong economical and statistical signifi-
cance of wage discrimination in the lower part of the salary distribution which
is aimed against Hispanic and African American players. Using the method
of ordinary least squares we found a wage differential for the average player of
2.37% and 5.35% that is accrued against black and Latin players. These values
were far from being statistically significant, though. Quantile regression reveals
that the premium is much higher for poorer players. When the lowest quar-
tile was the main focus of our observation, the wage gap against players from
Latin America reached 9.61%. Similarly, blacks suffer from salary inequality
that amounts to 11.31%. Once we looked at the bottom decile, the situa-
tion appeared to be more dramatical. Despite the same performance, black
![Page 57: An Empirical Investigation of Wage Discrimination in ... · jevili platovou diskriminaci v˚uˇci Afroameriˇcan˚um a hr´aˇc˚um z Latinsk´e Ameriky v nejniˇzˇs´ım decilu](https://reader030.vdocument.in/reader030/viewer/2022040907/5e7d12e667f4fa2fa646aac6/html5/thumbnails/57.jpg)
7. Conclusion 44
players receive 18.86% less. By implication, our thesis revealed that poorer
non-white players are more vulnerable as the employers pay them less despite
the economic inefficiency. The crucial lesson is that salary discrimination in
professional football exists in spite of the ample amount of supporters and per-
formance analysts. These findings were enabled especially thanks to the large
amount of observations that we have collected. Lately, significant results have
been rarely found, even though they exist, albeit only for a subset of the entire
population. Quantile regression is therefore a very powerful tool for detecting
discrimination, but the sample needs to be sufficiently large to observe differ-
entials at individual quantiles.
In the proposal, we stated 2 other research questions, which, given the
dataset we worked with, we were able to answer. With regards to a possi-
ble racial discrimination on different positions, we did not find any statistical
evidence. Even though the race plays an important role in determining which
position each footballer will play on, salaries are distributed justly according to
the player’s performance. The last question we wanted to clarify was whether
or not the increasing interest of fans and investors in MLS, and introduction
of new salary policies, have caused any wage gap. We used DID estimator to
find quite high, but insignificant effect of the increasing wage level on compen-
sations of black and Hispanic players.
The results of our thesis bring very surprising findings. The evidence of
racial discrimination in MLS might prompt others to examine and search for
racial discrimination in other sports areas. With regards to professional foot-
ball, the improvement of our analysis could be done, if the European best
competitions decided to publish yearly data about players’ salaries. In spite of
the increasing quality level in MLS, this league still lags behind the top leagues
from England, Spain or Germany where the salaries and differences between
them are incomparably higher (BBC 2015). However, for now, the reveal of
salary fees from European clubs is hardly expected as both clubs and players
are afraid of potential conflicts with tax authorities.
![Page 58: An Empirical Investigation of Wage Discrimination in ... · jevili platovou diskriminaci v˚uˇci Afroameriˇcan˚um a hr´aˇc˚um z Latinsk´e Ameriky v nejniˇzˇs´ım decilu](https://reader030.vdocument.in/reader030/viewer/2022040907/5e7d12e667f4fa2fa646aac6/html5/thumbnails/58.jpg)
Bibliography
Altonji, J. G. & C. R. Pierret (2001): “Employer learning and statistical
discrimination.” The Quarterly Journal of Economics 116(1): pp. 313–350.
Arrow, K. (1971): “The theory of discrimination. Princeton university, de-
partment of economics.” Industrial Relations Section pp. 313–350.
Arrow, K. J. (1998): “What has economics to say about racial discrimina-
tion?” The Journal of Economic Perspectives 12(2): pp. 91–100.
Ashenfelter, O. C. & D. Card (1984): “Using the longitudinal structure of
earnings to estimate the effect of training programs.”
BBC (2015): “How long would it take you to earn a top footballer’s salary?”
[online], Available at: http://www.bbc.com/news/world-31110113, [Ac-
cessed: July 12, 2017].
Becker, G. (1971): “The economics of discrimination 2nd ed. (University of
Chicago Press, Chicago).”
Blau, F. D. & L. M. Kahn (2016): “The gender wage gap: Extent, trends,
and explanations.” Technical report, National Bureau of Economic Research.
Brown, E., R. Spiro, & D. Keenan (1991): “Wage and nonwage discrimi-
nation in professional basketball: do fans affect it?” American Journal of
Economics and Sociology 50(3): pp. 333–345.
Cain, G. G. (1986): “The economic analysis of labor market discrimination:
A survey.” Handbook of Labor Economics 1: pp. 693–785.
Dunning, E. et al. (1999): “The development of soccer as a world game.”
Sport Matters: Sociological Studies of Sport, Violence and Civilisation. pp.
80–105.
![Page 59: An Empirical Investigation of Wage Discrimination in ... · jevili platovou diskriminaci v˚uˇci Afroameriˇcan˚um a hr´aˇc˚um z Latinsk´e Ameriky v nejniˇzˇs´ım decilu](https://reader030.vdocument.in/reader030/viewer/2022040907/5e7d12e667f4fa2fa646aac6/html5/thumbnails/59.jpg)
Bibliography 46
Eitzen, D. S., G. H. Sage et al. (1982): Sociology of American sport. Wm.
C. Brown Company Publishers.
Espnfc.com (2017): “MLS pumping money into salaries causes
ripple effect throughout league.” [online], Available at: http:
//www.espnfc.com/major-league-soccer/19/blog/post/3112526/
major-league-soccer-pumping-money-into-salaries-
causes-ripple-effect-throughout-league, [Accessed: June 12, 2017].
Fang, H. & A. Moro (2010): “Theories of statistical discrimination and af-
firmative action: A survey.” Technical report, National Bureau of Economic
Research.
Feinberg, W. (2003): “Affirmative action.” The Oxford Handbook of Practical
Ethics .
Fifa.com (2007): “265 million playing football.” [online], Available
at: https://www.fifa.com/mm/document/fifafacts/bcoffsurv/emaga_
9384_10704.pdf, [Accessed: March 25, 2017].
Forbes.com (2017): “MLS records banner year in 2016, cements
position among top u.s. pro sports leagues.” [online], Avail-
able at: https://www.forbes.com/sites/markjburns/2016/10/26/
mls-records-banner-year-in-2016-cements-position-among-top-us-
pro-sports-leagues/, [Accessed: June 12, 2017].
Frick, B. (2006): “Salary determination and the pay-performance relationship
in professional soccer: Evidence from Germany.” Sports Economics After
Fifty Years: Essays in Honour of Simon Rottenberg. Oviedo: Ediciones de
la Universidad de Oviedo pp. 125–146.
Fryer Jr, R. G. (2010): “Racial inequality in the 21st century: The de-
clining significance of discrimination.” Technical report, National Bureau of
Economic Research.
Gwartney, J. & C. Haworth (1974): “Employer costs and discrimination:
The case of baseball.” Journal of Political Economy 82(4): pp. 873–881.
Hamilton, B. H. (1997): “Racial discrimination and professional basketball
salaries in the 1990s.” Applied Economics 29(3): pp. 287–296.
![Page 60: An Empirical Investigation of Wage Discrimination in ... · jevili platovou diskriminaci v˚uˇci Afroameriˇcan˚um a hr´aˇc˚um z Latinsk´e Ameriky v nejniˇzˇs´ım decilu](https://reader030.vdocument.in/reader030/viewer/2022040907/5e7d12e667f4fa2fa646aac6/html5/thumbnails/60.jpg)
Bibliography 47
Hill, J. R. (2004): “Pay discrimination in the NBA revisited.” Quarterly
Journal of Business and Economics 43(1/2): pp. 81–92.
Holmes, P. (2011): “New evidence of salary discrimination in Major League
Baseball.” Labour Economics 18(3): pp. 320–331.
Jiobu, R. M. (1988): “Racial inequality in a public arena: The case of profes-
sional baseball.” Social Forces 67(2): pp. 524–534.
Kahn, L. M. (1991): “Discrimination in professional sports: A survey of the
literature.” ILR Review 44(3): pp. 395–418.
Kahn, L. M. & M. Shah (2005): “Race, compensation and contract length
in the NBA: 2001–2002.” Industrial Relations: A Journal of Economy and
Society 44(3): pp. 444–462.
Kahn, L. M. & P. D. Sherer (1988): “Racial differences in professional bas-
ketball players’ compensation.” Journal of Labor Economics 6(1): pp. 40–61.
Koenker, R. & G. Bassett Jr (1978): “Regression quantiles.” Economet-
rica: Journal of the Econometric Society pp. 33–50.
Koenker, R. & K. Hallock (2001): “Quantile regression: An introduction.”
Journal of Economic Perspectives 15(4): pp. 43–56.
Lang, K., J. Lehmann, & K. Yeon (2012): “Racial discrimination in the
labor market: Theory and empirics.” Journal of Economic Literature 50(4):
pp. 959–1006.
Lang, K. & M. Manove (2011): “Education and labor market discrimina-
tion.” The American Economic Review 101(4): pp. 1467–1496.
Lapchick (2016): “The 2016 racial and gender report card: Major League
Soccer.” [online], Available at: https://www.mlssoccer.com/glossary/
targeted-allocation-money, [Accessed: July 24, 2017].
Mlssoccer.com (1996): “1996 season recap.” [online], Available at: https:
//www.mlssoccer.com/history/season/1996, [Accessed: May 30, 2017].
Mlssoccer.com (2010): “MLS joins in FIFA Say No to Racism Day.”
[online], Available at: https://www.mlssoccer.com/post/2010/01/23/
mls-joins-fifa-say-no-racism-day, [Accessed: May 16, 2017].
![Page 61: An Empirical Investigation of Wage Discrimination in ... · jevili platovou diskriminaci v˚uˇci Afroameriˇcan˚um a hr´aˇc˚um z Latinsk´e Ameriky v nejniˇzˇs´ım decilu](https://reader030.vdocument.in/reader030/viewer/2022040907/5e7d12e667f4fa2fa646aac6/html5/thumbnails/61.jpg)
Bibliography 48
Mlssoccer.com (2017): “Targeted allocation money.” [online], Available at:
https://www.mlssoccer.com/glossary/targeted-allocation-money,
[Accessed: May 24, 2017].
Mueller, F., R. Cantu, & S. Van Camp (1996): “Team sports.” Catas-
trophic Injuries in High School and College Sports p. 57.
Nardinelli, C. & C. Simon (1990): “Customer racial discrimination in the
market for memorabilia: The case of baseball.” The Quarterly Journal of
Economics 105(3): pp. 575–595.
Oaxaca, R. (1973): “Male-female wage differentials in urban labor markets.”
International Economic Review pp. 693–709.
Oxford University (2017): “Definition of discrimination; origin.” [on-
line], Available at: https://en.oxforddictionaries.com/definition/
discriminate, [Accessed: March 11, 2017].
Pager, D., B. Bonikowski, & B. Western (2009): “Discrimination in a
low-wage labor market: A field experiment.” American Sociological Review
74(5): pp. 777–799.
Rehnstrom, K. (2009): “Racial salary discrimination in the NBA: 2008-
2009.” Major Themes in Economics 11: pp. 1–16.
Reuters (2017): “Beckham put MLS on fast track to respectabil-
ity.” [online], Available at: http://www.reuters.com/article/
us-soccer-usa-beckham-idUSKBN14U2E5, [Accessed: May 12, 2017].
Rosen, S. & A. Sanderson (2001): “Labour markets in professional sports.”
The Economic Journal 111(469): pp. 47–68.
Scully, G. W. (1973): “Economic discrimination in professional sports.” Law
and Contemporary Problems 38(1): pp. 67–84.
Scully, G. W. (1974): “Pay and performance in Major League Baseball.” The
American Economic Review 64(6): pp. 915–930.
Sears, D. O. (1988): “Symbolic racism.” In “Eliminating racism,” pp. 53–84.
Springer.
![Page 62: An Empirical Investigation of Wage Discrimination in ... · jevili platovou diskriminaci v˚uˇci Afroameriˇcan˚um a hr´aˇc˚um z Latinsk´e Ameriky v nejniˇzˇs´ım decilu](https://reader030.vdocument.in/reader030/viewer/2022040907/5e7d12e667f4fa2fa646aac6/html5/thumbnails/62.jpg)
Bibliography 49
Sommers, P. M. & N. Quinton (1982): “Pay and performance in Major
League Baseball: The case of the first family of free agents.” The Journal of
Human Resources 17(3): pp. 426–436.
Sportingintelligence.com (2016): “Global sports salaries survey 2016.”
[online], Available at: https://www.globalsportssalaries.com/GSSS%
202016.pdf, [Accessed: March 14, 2017].
Statista.com (2017): “Average per game attendance of the five
major sports leagues in North America 2016/17.” [online],
Available at: https://www.statista.com/statistics/207458/
per-game-attendance-of-major-us-sports-leagues/, [Accessed:
June 18, 2017].
Szymanski, S. (2000): “A market test for discrimination in the English profes-
sional soccer leagues.” Journal of Political Economy 108(3): pp. 590–603.
Takaki, R. T. (1979): Iron Cages: Race and Culture in 19th-century America.
Oxford University Press, USA.
Wells, L. T., N. Allen, J. Morisset, & N. Pirnia (2001): Using Tax
Incentives to Compete for Foreign Investment: Are They Worth the Cost?
Washington, DC: FIAS.
Wilson, D. P. & Y.-H. Ying (2003): “Nationality preferences for labour in the
international football industry.” Applied Economics 35(14): pp. 1551–1559.
Wooldridge, J. M. (2015): Introductory Econometrics: A Modern Approach.
Nelson Education.
Yang, C.-H. & H.-Y. Lin (2012): “Is there salary discrimination by nationality
in the NBA? foreign talent or foreign market.” Journal of Sports Economics
13(1): pp. 53–75.
Zimbalist, A. (2003): “Sport as business.” Oxford Review of Economic Policy
19(4): pp. 503–511.
![Page 63: An Empirical Investigation of Wage Discrimination in ... · jevili platovou diskriminaci v˚uˇci Afroameriˇcan˚um a hr´aˇc˚um z Latinsk´e Ameriky v nejniˇzˇs´ım decilu](https://reader030.vdocument.in/reader030/viewer/2022040907/5e7d12e667f4fa2fa646aac6/html5/thumbnails/63.jpg)
Appendix A
OLS Assumptions
1. Linear in Parameters: In the population model, the relation between
the dependent variable and the independent variables is assumed to be
linear. Therefore, ~α, ~β and ~γ are linear vectors.
2. Random Sampling: Our sample consists of all players who satisfy given
condition. Once this condition was satisfied, we measured characteristics
of all players to use this information and make an inference about the en-
tire population. To prove the randomness of our sample, we compared the
racial composition with the data reported by the Institute for Diversity
and Ethnics in Sport. Values of 48.8% (whites) and 24.8% (Latinos) from
our dataset were almost perfectly matched by 48% and 24.8% (Lapchick
2016). The proportion of black players is split into national and interna-
tional, but as the percentage of Asians is only 0.7%, their representation
in our sample will be very similar as well.
3. No Perfect Collinearity: In our sample, there is no variable that is
constant, and there are no perfect linear relationships among explanatory
variables included into the model. If it was so, our matrix X of explana-
tory variables would be singular and it would have perfect multilinear
dependence. In this case, we would not be able to compute OLS estima-
tor since matrix (X’X) could not be inverted. Table A.1 depicts mutual
correlations between independent variables and most of the values show
no signs of perfect collinearity. Very strong collinearity is between Age
and Age2, but the inclusion of both of them was necessary due to the
non-linear relationship between the player’s age and salary.
4. No serial correlation: This assumption rules out the correlation of
![Page 64: An Empirical Investigation of Wage Discrimination in ... · jevili platovou diskriminaci v˚uˇci Afroameriˇcan˚um a hr´aˇc˚um z Latinsk´e Ameriky v nejniˇzˇs´ım decilu](https://reader030.vdocument.in/reader030/viewer/2022040907/5e7d12e667f4fa2fa646aac6/html5/thumbnails/64.jpg)
A. OLS Assumptions II
Table A.1: Correlation of explanatory variables
Variable Str Midf Star SOG GS Black Hisp MINS G A SHTS FoCo FoSu Offs Yc Age Age2
Str 1 -0.79 -0.05 0.38 -0.28 0.13 0.02 -0.28 0.44 -0.13 0.31 -0.11 -0.04 0.56 -0.28 -0.01 -0.01Midf -0.79 1 0.02 -0.15 0.2 -0.14 0.05 0.19 -0.25 0.28 -0.08 0.09 0.15 -0.39 0.18 0.02 0.02Star -0.05 0.02 1 0.16 0.22 -0.01 -0.12 0.24 0.17 0.2 0.14 0.13 0.21 0.04 0.13 0.18 0.18SOG 0.38 -0.15 0.16 1 0.38 0.01 0.09 0.38 0.88 0.41 0.95 0.23 0.41 0.68 0.02 0.13 0.13GS -0.28 0.2 0.22 0.38 1 -0.12 0.05 0.99 0.29 0.45 0.44 0.51 0.52 0.1 0.41 0.2 0.19Black 0.13 -0.14 -0.01 0.01 -0.12 1 -0.34 -0.11 0.05 -0.15 0 0 -0.01 0.11 -0.11 -0.11 -0.11Hisp 0.02 0.05 -0.12 0.09 0.05 -0.34 1 0.03 0.09 0.15 0.09 0.09 0.15 0.05 0.13 0.08 0.09MINS -0.28 0.19 0.24 0.38 0.99 -0.11 0.03 1 0.3 0.45 0.44 0.52 0.52 0.1 0.42 0.19 0.17G 0.44 -0.25 0.17 0.88 0.29 0.05 0.09 0.3 1 0.32 0.82 0.16 0.31 0.67 -0.04 0.2 0.20A -0.13 0.28 0.2 0.41 0.45 -0.15 0.15 0.45 0.32 1 0.43 0.18 0.48 0.11 0.13 0.26 0.26SHTS 0.31 -0.08 0.14 0.95 0.44 0 0.09 0.44 0.82 0.43 1 0.27 0.44 0.62 0.07 0.12 0.12FoCo -0.11 0.09 0.13 0.23 0.51 0 0.09 0.52 0.16 0.18 0.27 1 0.5 0.12 0.63 0.07 0.05FoSu -0.04 0.15 0.21 0.41 0.52 -0.01 0.15 0.52 0.31 0.48 0.44 0.5 1 0.21 0.31 0.11 0.10Offs 0.56 -0.39 0.04 0.68 0.1 0.11 0.05 0.1 0.67 0.11 0.62 0.12 0.21 1 -0.06 0.17 0.17Yc -0.28 0.18 0.13 0.02 0.41 -0.11 0.13 0.42 -0.04 0.13 0.07 0.63 0.31 -0.06 1 0.14 0.127Age -0.01 0.02 0.18 0.13 0.2 -0.11 0.08 0.19 0.2 0.26 0.12 0.07 0.11 0.17 0.14 1 0.996Age2 -0.01 0.02 0.18 0.13 0.19 -0.11 0.09 0.17 0.2 0.26 0.12 0.05 0.1 0.17 0.13 0.996 1
errors from two different time periods. This is not a problem for our data
as we did not observe the same individuals for each time period.
5. Homoskedasticity: This assumption ensures that the error u has con-
stant variance for all values of independent variables. The most common
tests for homoskedasticity are White’s test and Breusch-Pagan. The B-P
test uncovers heteroskedasticity and hence, we also present White’s stan-
dard errors that are heteroscedasticity-consistent. The results show that
the majority of explanatory variables preserve their statistical significance
(see Appendix B). We reject the null hypothesis of homoskedasticity at
a very low p-value.
Studentized Breusch-Pagan test: H0: Homoskedasticity
BP = 127.0637, df = 27, p-value = 6.396× 10−7
6. Normality: The errors u are not dependent on explanatory variables
and are normally distributed with zero mean and variance equal to σ2 for
the whole population: u ∼N(0,σ2). There are two common hypotheses
about how to check this assumption to be valid - The Anderson-Darling
test and the Shapiro-Wilk test. The null hypothesis H0 is that the data
is normally distributed for both of them. Running the S-W test we reject
the null hypothesis of normality.
![Page 65: An Empirical Investigation of Wage Discrimination in ... · jevili platovou diskriminaci v˚uˇci Afroameriˇcan˚um a hr´aˇc˚um z Latinsk´e Ameriky v nejniˇzˇs´ım decilu](https://reader030.vdocument.in/reader030/viewer/2022040907/5e7d12e667f4fa2fa646aac6/html5/thumbnails/65.jpg)
A. OLS Assumptions III
Shapiro-Wilk normality test: H0: Normality of errors
W = 0.9719, p-value = 8.868× 10−6
From the figure A.1 we can observe that the errors are right skewed,
meaning that most of the error terms are with a long tail on the left
side of the distribution. The shape of the Q-Q plot is heavily influenced
by the extreme values in the highest decile of the salary distribution.
Removing these outliers would cause a loss of a valuable information,
though. Even though our sample consists of 1,100 observations, relying
on Central Limit Theorem may not be appropriate due to long-tailed
errors. Consequently, we also run a robust regression (see Appendix B)
that weights and dampens the effect of outliers to find out that both
results and statistical significances1 are very similar and inference about
discrimination is not influenced. As both assumptions are not required
for the quantile regression, we can rely on the discriminating behaviour
we have uncovered in the lower part of the salary distribution.
Figure A.1: Q-Q plot of the regression
1The only variable that differs across different method for 3-seasons statistics is the posi-tion of striker. Its negative coefficient is different from zero using robust regression.
![Page 66: An Empirical Investigation of Wage Discrimination in ... · jevili platovou diskriminaci v˚uˇci Afroameriˇcan˚um a hr´aˇc˚um z Latinsk´e Ameriky v nejniˇzˇs´ım decilu](https://reader030.vdocument.in/reader030/viewer/2022040907/5e7d12e667f4fa2fa646aac6/html5/thumbnails/66.jpg)
Appendix B
Robust methods
Table B.1: Comparison between OLS and White’s standard errors
1-season statistics 3-seasons statistics
Variable Estimate Robust SE OLS SE Stat. Significance Estimate Robust SE OLS SE Stat. Significance
Constant 12.255 0.984 0.811 *** 12.127 0.949 0.796 ***STRIKER -0.087 0.090 0.098 - -0.152 0.090 0.099 */-MIDFIELD -0.125 0.072 0.085 * -0.176 0.072 0.085 **GS 0.035 0.019 0.020 * 0.063 0.024 0.024 **MINS -0.001 0.0002 0.0002 ** -0.001 0.0003 0.0003 ***GOALS 0.043 0.014 0.012 *** 0.059 0.018 0.015 ***ASSISTS 0.032 0.011 0.009 *** 0.045 0.013 0.011 ***SHTS 0.009 0.003 0.003 *** 0.007 0.004 0.004 *SOG -0.007 0.008 0.008 - 0.005 0.011 0.011 -FOCO -0.006 0.003 0.002 */** -0.008 0.003 0.003 **/***FOSU 0.006 0.002 0.001 *** 0.004 0.002 0.002 *YCARD 0.028 0.015 0.014 */** 0.066 0.018 0.017 ***OFFS -0.005 0.003 0.003 - -0.009 0.004 0.004 **STAR 0.730 0.091 0.073 *** 0.571 0.084 0.069 ***AGE -0.169 0.076 0.060 **/*** -0.160 0.073 0.059 **/***AGESQ 0.005 0.001 0.001 *** 0.005 0.001 0.001 ***BLACK -0.045 0.060 0.059 - -0.024 0.059 0.058 -HISP -0.029 0.066 0.061 - -0.055 0.064 0.060 -year2016 0.770 0.112 0.117 *** 0.774 0.117 0.116 ***year2015 0.887 0.118 0.115 *** 0.854 0.124 0.115 ***year2014 0.557 0.110 0.116 *** 0.581 0.114 0.116 ***year2013 0.495 0.101 0.116 *** 0.551 0.105 0.115 ***year2012 0.499 0.109 0.113 *** 0.549 0.110 0.112 ***year2011 0.302 0.111 0.114 *** 0.366 0.111 0.111 ***year2010 0.287 0.110 0.113 ***/** 0.297 0.106 0.110 ***year2009 0.168 0.103 0.112 - 0.135 0.102 0.109 -year2008 0.021 0.106 0.111 - 0.035 0.106 0.108 -year2007 0.062 0.103 0.110 - 0.001 0.103 0.107 -
*Significant at 10% level, **Significant at 5% level, ***Significant at 1% level
![Page 67: An Empirical Investigation of Wage Discrimination in ... · jevili platovou diskriminaci v˚uˇci Afroameriˇcan˚um a hr´aˇc˚um z Latinsk´e Ameriky v nejniˇzˇs´ım decilu](https://reader030.vdocument.in/reader030/viewer/2022040907/5e7d12e667f4fa2fa646aac6/html5/thumbnails/67.jpg)
B. Robust methods V
Table B.2: Comparison between OLS and Robust regressions
1-season statistics 3-seasons statistics
Variable OLS OLS SE Robust Robust SE Stat. Significance OLS OLS SE Robust Robust SE Stat. Significance
Constant 12.255 0.811 12.051 0.749 *** 12.127 0.796 12.071 0.735 ***STRIKER -0.087 0.098 -0.09 0.09 - -0.152 0.099 -0.189 0.091 -/**MIDFIELD -0.125 0.085 -0.111 0.079 */- -0.176 0.085 -0.191 0.079 **GS 0.035 0.02 0.024 0.018 */- 0.063 0.024 0.051 0.022 **MINS -0.0007 0.0003 -0.0004 0.0002 **/* -0.001 0.0003 -0.0007 0.0003 ***GOALS 0.043 0.012 0.041 0.011 *** 0.059 0.015 0.061 0.015 ***ASSISTS 0.032 0.009 0.021 0.008 ***/** 0.045 0.011 0.044 0.011 ***SHTS 0.009 0.003 0.01 0.003 *** 0.007 0.004 0.008 0.004 */**SOG -0.007 0.008 -0.010 0.008 - 0.005 0.011 0.0002 0.01 -FOCO -0.006 0.002 -0.005 0.002 ** -0.008 0.003 -0.007 0.003 ***/**FOSU 0.006 0.001 0.005 0.002 *** 0.004 0.002 0.004 0.002 */**YCARD 0.028 0.014 0.022 0.013 **/* 0.066 0.017 0.059 0.016 ***OFFS -0.005 0.003 -0.004 0.003 - -0.009 0.004 -0.007 0.004 **/*STAR 0.730 0.073 0.719 0.068 *** 0.571 0.069 0.478 0.064 ***AGE -0.169 0.06 -0.158 0.056 *** -0.16 0.059 -0.163 0.055 ***AGESQ 0.005 0.001 0.005 0.001 *** 0.005 0.001 0.005 0.001 ***BLACK -0.045 0.059 -0.054 0.055 - -0.024 0.058 -0.040 0.054 -HISP -0.029 0.061 0.007 0.056 - -0.055 0.060 -0.053 0.056 -year2016 0.770 0.117 0.790 0.108 *** 0.774 0.116 0.776 0.107 ***year2015 0.887 0.115 0.844 0.106 *** 0.854 0.115 0.800 0.107 ***year2014 0.557 0.116 0.576 0.107 *** 0.581 0.116 0.578 0.107 ***year2013 0.495 0.116 0.506 0.107 *** 0.551 0.115 0.542 0.106 ***year2012 0.499 0.113 0.536 0.105 *** 0.549 0.112 0.554 0.104 ***year2011 0.302 0.114 0.306 0.105 *** 0.366 0.111 0.394 0.103 ***year2010 0.287 0.113 0.322 0.104 **/*** 0.297 0.110 0.342 0.102 ***year2009 0.168 0.112 0.218 0.103 -/** 0.135 0.109 0.187 0.101 -year2008 0.021 0.111 0.075 0.102 - 0.035 0.108 0.081 0.100 -year2007 0.062 0.110 0.101 0.102 - 0.001 0.107 0.039 0.099 -
*Significant at 10% level, **Significant at 5% level, ***Significant at 1% level
Note: The method we have used is called Huber’s M-estimator and it deadens the effect ofoutliers.
![Page 68: An Empirical Investigation of Wage Discrimination in ... · jevili platovou diskriminaci v˚uˇci Afroameriˇcan˚um a hr´aˇc˚um z Latinsk´e Ameriky v nejniˇzˇs´ım decilu](https://reader030.vdocument.in/reader030/viewer/2022040907/5e7d12e667f4fa2fa646aac6/html5/thumbnails/68.jpg)
Appendix C
Unrestricted vs. restricted model
Table C.1: Results after the exclusion of insignificant variables
Variable Before (t-stats) After (t-stats) Difference Percentage change
Constant 12.127*** (-15.24) 11.884*** (15.231) 0.243 27.51%MIDFIELD -0.176** (-2.053) -0.077 (-1.385) -0.099 -9.43%GS 0.063*** (-2.653) 0.062*** (2.662) 0.001 0.1%MINS -0.001*** (-3.141) -0.001*** (-2.98) 0 0%GOALS 0.059*** (-3.733) 0.06*** (4.645) -0.001 -0.1%ASSISTS 0.045*** (-3.975) 0.044*** (3.953) 0.001 0.1%SHTS 0.007* (-1.85) 0.008*** (3.757) -0.001 -0.1%FOCO -0.008*** (-2.691) -0.009** (-3.088) 0.001 0.1%FOSU 0.004* (-1.905) 0.003 (1.546) 0.001 0.1%YCARD 0.066*** (-3.842) 0.069*** (4.144) -0.003 -0.3%OFFS -0.009** (-2.150) -0.01** (-2.521) 0.001 0.1%STAR 0.571*** (-8.226) 0.583*** (8.512) -0.012 -1.19%AGE -0.160*** (-2.687) -0.148** (-2.517) -0.012 -1.19%AGE2 0.005*** (-4.241) 0.004*** (4.092) 0.001 0.1%year2016 0.774*** (-6.677) 0.708*** (7.96) 0.066 6.82%year2015 0.854*** (-7.408) 0.783*** (8.844) 0.071 7.36%year2014 0.581*** (-5.015) 0.507*** (5.773) 0.074 7.68%year2013 0.551*** (-4.785) 0.48*** (5.562) 0.071 7.36%year2012 0.549*** (-4.888) 0.484*** (5.618) 0.065 6.72%year2011 0.366*** (-3.279) 0.301*** (3.51) 0.065 6.72%year2010 0.297*** (-2.689) 0.239** (2.788) 0.058 5.97%
R2 0.532 0.530Adjusted R2 0.520 0.521N 1100 1100F-statistic (DF) 45.11 (27; 1072) 60.76 (20; 1079)
*Significant at 10% level, **Significant at 5% level, ***Significant at 1% level